blob: b2feda35966b124652c662f65392231ca464875c [file] [log] [blame]
Vishal Verma5212e112015-06-25 04:20:32 -04001/*
2 * Block Translation Table
3 * Copyright (c) 2014-2015, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 */
14#include <linux/highmem.h>
15#include <linux/debugfs.h>
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/device.h>
19#include <linux/mutex.h>
20#include <linux/hdreg.h>
21#include <linux/genhd.h>
22#include <linux/sizes.h>
23#include <linux/ndctl.h>
24#include <linux/fs.h>
25#include <linux/nd.h>
26#include "btt.h"
27#include "nd.h"
28
29enum log_ent_request {
30 LOG_NEW_ENT = 0,
31 LOG_OLD_ENT
32};
33
Vishal Verma86652d22017-09-05 14:35:39 -060034static struct device *to_dev(struct arena_info *arena)
35{
36 return &arena->nd_btt->dev;
37}
38
Vishal Vermad9b83c72017-08-30 19:36:03 -060039static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset)
40{
41 return offset + nd_btt->initial_offset;
42}
43
Vishal Verma5212e112015-06-25 04:20:32 -040044static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
Vishal Verma3ae3d672017-05-10 15:01:30 -060045 void *buf, size_t n, unsigned long flags)
Vishal Verma5212e112015-06-25 04:20:32 -040046{
47 struct nd_btt *nd_btt = arena->nd_btt;
48 struct nd_namespace_common *ndns = nd_btt->ndns;
49
Vishal Verma14e49452017-06-28 14:25:00 -060050 /* arena offsets may be shifted from the base of the device */
Vishal Vermad9b83c72017-08-30 19:36:03 -060051 offset = adjust_initial_offset(nd_btt, offset);
Vishal Verma3ae3d672017-05-10 15:01:30 -060052 return nvdimm_read_bytes(ndns, offset, buf, n, flags);
Vishal Verma5212e112015-06-25 04:20:32 -040053}
54
55static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
Vishal Verma3ae3d672017-05-10 15:01:30 -060056 void *buf, size_t n, unsigned long flags)
Vishal Verma5212e112015-06-25 04:20:32 -040057{
58 struct nd_btt *nd_btt = arena->nd_btt;
59 struct nd_namespace_common *ndns = nd_btt->ndns;
60
Vishal Verma14e49452017-06-28 14:25:00 -060061 /* arena offsets may be shifted from the base of the device */
Vishal Vermad9b83c72017-08-30 19:36:03 -060062 offset = adjust_initial_offset(nd_btt, offset);
Vishal Verma3ae3d672017-05-10 15:01:30 -060063 return nvdimm_write_bytes(ndns, offset, buf, n, flags);
Vishal Verma5212e112015-06-25 04:20:32 -040064}
65
66static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
67{
68 int ret;
69
Vishal Vermab177fe82017-05-10 15:01:31 -060070 /*
71 * infooff and info2off should always be at least 512B aligned.
72 * We rely on that to make sure rw_bytes does error clearing
73 * correctly, so make sure that is the case.
74 */
Vishal Verma86652d22017-09-05 14:35:39 -060075 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512),
76 "arena->infooff: %#llx is unaligned\n", arena->infooff);
77 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512),
78 "arena->info2off: %#llx is unaligned\n", arena->info2off);
Vishal Vermab177fe82017-05-10 15:01:31 -060079
Vishal Verma5212e112015-06-25 04:20:32 -040080 ret = arena_write_bytes(arena, arena->info2off, super,
Vishal Verma3ae3d672017-05-10 15:01:30 -060081 sizeof(struct btt_sb), 0);
Vishal Verma5212e112015-06-25 04:20:32 -040082 if (ret)
83 return ret;
84
85 return arena_write_bytes(arena, arena->infooff, super,
Vishal Verma3ae3d672017-05-10 15:01:30 -060086 sizeof(struct btt_sb), 0);
Vishal Verma5212e112015-06-25 04:20:32 -040087}
88
89static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
90{
Vishal Verma5212e112015-06-25 04:20:32 -040091 return arena_read_bytes(arena, arena->infooff, super,
Vishal Verma3ae3d672017-05-10 15:01:30 -060092 sizeof(struct btt_sb), 0);
Vishal Verma5212e112015-06-25 04:20:32 -040093}
94
95/*
96 * 'raw' version of btt_map write
97 * Assumptions:
98 * mapping is in little-endian
99 * mapping contains 'E' and 'Z' flags as desired
100 */
Vishal Verma3ae3d672017-05-10 15:01:30 -0600101static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping,
102 unsigned long flags)
Vishal Verma5212e112015-06-25 04:20:32 -0400103{
104 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
105
Vishal Verma86652d22017-09-05 14:35:39 -0600106 if (unlikely(lba >= arena->external_nlba))
107 dev_err_ratelimited(to_dev(arena),
108 "%s: lba %#x out of range (max: %#x)\n",
109 __func__, lba, arena->external_nlba);
Vishal Verma3ae3d672017-05-10 15:01:30 -0600110 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags);
Vishal Verma5212e112015-06-25 04:20:32 -0400111}
112
113static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
Vishal Verma3ae3d672017-05-10 15:01:30 -0600114 u32 z_flag, u32 e_flag, unsigned long rwb_flags)
Vishal Verma5212e112015-06-25 04:20:32 -0400115{
116 u32 ze;
117 __le32 mapping_le;
118
119 /*
120 * This 'mapping' is supposed to be just the LBA mapping, without
121 * any flags set, so strip the flag bits.
122 */
Vishal Verma0595d532017-08-30 19:35:59 -0600123 mapping = ent_lba(mapping);
Vishal Verma5212e112015-06-25 04:20:32 -0400124
125 ze = (z_flag << 1) + e_flag;
126 switch (ze) {
127 case 0:
128 /*
129 * We want to set neither of the Z or E flags, and
130 * in the actual layout, this means setting the bit
131 * positions of both to '1' to indicate a 'normal'
132 * map entry
133 */
134 mapping |= MAP_ENT_NORMAL;
135 break;
136 case 1:
137 mapping |= (1 << MAP_ERR_SHIFT);
138 break;
139 case 2:
140 mapping |= (1 << MAP_TRIM_SHIFT);
141 break;
142 default:
143 /*
144 * The case where Z and E are both sent in as '1' could be
145 * construed as a valid 'normal' case, but we decide not to,
146 * to avoid confusion
147 */
Vishal Verma86652d22017-09-05 14:35:39 -0600148 dev_err_ratelimited(to_dev(arena),
149 "Invalid use of Z and E flags\n");
Vishal Verma5212e112015-06-25 04:20:32 -0400150 return -EIO;
151 }
152
153 mapping_le = cpu_to_le32(mapping);
Vishal Verma3ae3d672017-05-10 15:01:30 -0600154 return __btt_map_write(arena, lba, mapping_le, rwb_flags);
Vishal Verma5212e112015-06-25 04:20:32 -0400155}
156
157static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
Vishal Verma3ae3d672017-05-10 15:01:30 -0600158 int *trim, int *error, unsigned long rwb_flags)
Vishal Verma5212e112015-06-25 04:20:32 -0400159{
160 int ret;
161 __le32 in;
162 u32 raw_mapping, postmap, ze, z_flag, e_flag;
163 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
164
Vishal Verma86652d22017-09-05 14:35:39 -0600165 if (unlikely(lba >= arena->external_nlba))
166 dev_err_ratelimited(to_dev(arena),
167 "%s: lba %#x out of range (max: %#x)\n",
168 __func__, lba, arena->external_nlba);
Vishal Verma5212e112015-06-25 04:20:32 -0400169
Vishal Verma3ae3d672017-05-10 15:01:30 -0600170 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags);
Vishal Verma5212e112015-06-25 04:20:32 -0400171 if (ret)
172 return ret;
173
174 raw_mapping = le32_to_cpu(in);
175
Vishal Verma0595d532017-08-30 19:35:59 -0600176 z_flag = ent_z_flag(raw_mapping);
177 e_flag = ent_e_flag(raw_mapping);
Vishal Verma5212e112015-06-25 04:20:32 -0400178 ze = (z_flag << 1) + e_flag;
Vishal Verma0595d532017-08-30 19:35:59 -0600179 postmap = ent_lba(raw_mapping);
Vishal Verma5212e112015-06-25 04:20:32 -0400180
181 /* Reuse the {z,e}_flag variables for *trim and *error */
182 z_flag = 0;
183 e_flag = 0;
184
185 switch (ze) {
186 case 0:
187 /* Initial state. Return postmap = premap */
188 *mapping = lba;
189 break;
190 case 1:
191 *mapping = postmap;
192 e_flag = 1;
193 break;
194 case 2:
195 *mapping = postmap;
196 z_flag = 1;
197 break;
198 case 3:
199 *mapping = postmap;
200 break;
201 default:
202 return -EIO;
203 }
204
205 if (trim)
206 *trim = z_flag;
207 if (error)
208 *error = e_flag;
209
210 return ret;
211}
212
Vishal Verma166f39b2017-12-18 09:28:39 -0700213static int btt_log_group_read(struct arena_info *arena, u32 lane,
214 struct log_group *log)
Vishal Verma5212e112015-06-25 04:20:32 -0400215{
Vishal Verma5212e112015-06-25 04:20:32 -0400216 return arena_read_bytes(arena,
Vishal Verma166f39b2017-12-18 09:28:39 -0700217 arena->logoff + (lane * LOG_GRP_SIZE), log,
218 LOG_GRP_SIZE, 0);
Vishal Verma5212e112015-06-25 04:20:32 -0400219}
220
221static struct dentry *debugfs_root;
222
223static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
224 int idx)
225{
226 char dirname[32];
227 struct dentry *d;
228
229 /* If for some reason, parent bttN was not created, exit */
230 if (!parent)
231 return;
232
233 snprintf(dirname, 32, "arena%d", idx);
234 d = debugfs_create_dir(dirname, parent);
235 if (IS_ERR_OR_NULL(d))
236 return;
237 a->debugfs_dir = d;
238
239 debugfs_create_x64("size", S_IRUGO, d, &a->size);
240 debugfs_create_x64("external_lba_start", S_IRUGO, d,
241 &a->external_lba_start);
242 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
243 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
244 &a->internal_lbasize);
245 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
246 debugfs_create_u32("external_lbasize", S_IRUGO, d,
247 &a->external_lbasize);
248 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
249 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
250 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
251 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
252 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
253 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
254 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
255 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
256 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
257 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
Vishal Verma166f39b2017-12-18 09:28:39 -0700258 debugfs_create_u32("log_index_0", S_IRUGO, d, &a->log_index[0]);
259 debugfs_create_u32("log_index_1", S_IRUGO, d, &a->log_index[1]);
Vishal Verma5212e112015-06-25 04:20:32 -0400260}
261
262static void btt_debugfs_init(struct btt *btt)
263{
264 int i = 0;
265 struct arena_info *arena;
266
267 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
268 debugfs_root);
269 if (IS_ERR_OR_NULL(btt->debugfs_dir))
270 return;
271
272 list_for_each_entry(arena, &btt->arena_list, list) {
273 arena_debugfs_init(arena, btt->debugfs_dir, i);
274 i++;
275 }
276}
277
Vishal Verma166f39b2017-12-18 09:28:39 -0700278static u32 log_seq(struct log_group *log, int log_idx)
279{
280 return le32_to_cpu(log->ent[log_idx].seq);
281}
282
Vishal Verma5212e112015-06-25 04:20:32 -0400283/*
284 * This function accepts two log entries, and uses the
285 * sequence number to find the 'older' entry.
286 * It also updates the sequence number in this old entry to
287 * make it the 'new' one if the mark_flag is set.
288 * Finally, it returns which of the entries was the older one.
289 *
290 * TODO The logic feels a bit kludge-y. make it better..
291 */
Vishal Verma166f39b2017-12-18 09:28:39 -0700292static int btt_log_get_old(struct arena_info *a, struct log_group *log)
Vishal Verma5212e112015-06-25 04:20:32 -0400293{
Vishal Verma166f39b2017-12-18 09:28:39 -0700294 int idx0 = a->log_index[0];
295 int idx1 = a->log_index[1];
Vishal Verma5212e112015-06-25 04:20:32 -0400296 int old;
297
298 /*
299 * the first ever time this is seen, the entry goes into [0]
300 * the next time, the following logic works out to put this
301 * (next) entry into [1]
302 */
Vishal Verma166f39b2017-12-18 09:28:39 -0700303 if (log_seq(log, idx0) == 0) {
304 log->ent[idx0].seq = cpu_to_le32(1);
Vishal Verma5212e112015-06-25 04:20:32 -0400305 return 0;
306 }
307
Vishal Verma166f39b2017-12-18 09:28:39 -0700308 if (log_seq(log, idx0) == log_seq(log, idx1))
Vishal Verma5212e112015-06-25 04:20:32 -0400309 return -EINVAL;
Vishal Verma166f39b2017-12-18 09:28:39 -0700310 if (log_seq(log, idx0) + log_seq(log, idx1) > 5)
Vishal Verma5212e112015-06-25 04:20:32 -0400311 return -EINVAL;
312
Vishal Verma166f39b2017-12-18 09:28:39 -0700313 if (log_seq(log, idx0) < log_seq(log, idx1)) {
314 if ((log_seq(log, idx1) - log_seq(log, idx0)) == 1)
Vishal Verma5212e112015-06-25 04:20:32 -0400315 old = 0;
316 else
317 old = 1;
318 } else {
Vishal Verma166f39b2017-12-18 09:28:39 -0700319 if ((log_seq(log, idx0) - log_seq(log, idx1)) == 1)
Vishal Verma5212e112015-06-25 04:20:32 -0400320 old = 1;
321 else
322 old = 0;
323 }
324
325 return old;
326}
327
Vishal Verma5212e112015-06-25 04:20:32 -0400328/*
329 * This function copies the desired (old/new) log entry into ent if
330 * it is not NULL. It returns the sub-slot number (0 or 1)
331 * where the desired log entry was found. Negative return values
332 * indicate errors.
333 */
334static int btt_log_read(struct arena_info *arena, u32 lane,
335 struct log_entry *ent, int old_flag)
336{
337 int ret;
338 int old_ent, ret_ent;
Vishal Verma166f39b2017-12-18 09:28:39 -0700339 struct log_group log;
Vishal Verma5212e112015-06-25 04:20:32 -0400340
Vishal Verma166f39b2017-12-18 09:28:39 -0700341 ret = btt_log_group_read(arena, lane, &log);
Vishal Verma5212e112015-06-25 04:20:32 -0400342 if (ret)
343 return -EIO;
344
Vishal Verma166f39b2017-12-18 09:28:39 -0700345 old_ent = btt_log_get_old(arena, &log);
Vishal Verma5212e112015-06-25 04:20:32 -0400346 if (old_ent < 0 || old_ent > 1) {
Vishal Vermae6be2dc2017-06-30 18:32:51 -0600347 dev_err(to_dev(arena),
Vishal Verma5212e112015-06-25 04:20:32 -0400348 "log corruption (%d): lane %d seq [%d, %d]\n",
Vishal Verma166f39b2017-12-18 09:28:39 -0700349 old_ent, lane, log.ent[arena->log_index[0]].seq,
350 log.ent[arena->log_index[1]].seq);
Vishal Verma5212e112015-06-25 04:20:32 -0400351 /* TODO set error state? */
352 return -EIO;
353 }
354
355 ret_ent = (old_flag ? old_ent : (1 - old_ent));
356
357 if (ent != NULL)
Vishal Verma166f39b2017-12-18 09:28:39 -0700358 memcpy(ent, &log.ent[arena->log_index[ret_ent]], LOG_ENT_SIZE);
Vishal Verma5212e112015-06-25 04:20:32 -0400359
360 return ret_ent;
361}
362
363/*
364 * This function commits a log entry to media
365 * It does _not_ prepare the freelist entry for the next write
366 * btt_flog_write is the wrapper for updating the freelist elements
367 */
368static int __btt_log_write(struct arena_info *arena, u32 lane,
Vishal Verma3ae3d672017-05-10 15:01:30 -0600369 u32 sub, struct log_entry *ent, unsigned long flags)
Vishal Verma5212e112015-06-25 04:20:32 -0400370{
371 int ret;
Vishal Verma166f39b2017-12-18 09:28:39 -0700372 u32 group_slot = arena->log_index[sub];
373 unsigned int log_half = LOG_ENT_SIZE / 2;
Vishal Verma5212e112015-06-25 04:20:32 -0400374 void *src = ent;
Vishal Verma166f39b2017-12-18 09:28:39 -0700375 u64 ns_off;
Vishal Verma5212e112015-06-25 04:20:32 -0400376
Vishal Verma166f39b2017-12-18 09:28:39 -0700377 ns_off = arena->logoff + (lane * LOG_GRP_SIZE) +
378 (group_slot * LOG_ENT_SIZE);
Vishal Verma5212e112015-06-25 04:20:32 -0400379 /* split the 16B write into atomic, durable halves */
Vishal Verma3ae3d672017-05-10 15:01:30 -0600380 ret = arena_write_bytes(arena, ns_off, src, log_half, flags);
Vishal Verma5212e112015-06-25 04:20:32 -0400381 if (ret)
382 return ret;
383
384 ns_off += log_half;
385 src += log_half;
Vishal Verma3ae3d672017-05-10 15:01:30 -0600386 return arena_write_bytes(arena, ns_off, src, log_half, flags);
Vishal Verma5212e112015-06-25 04:20:32 -0400387}
388
389static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub,
390 struct log_entry *ent)
391{
392 int ret;
393
Vishal Verma3ae3d672017-05-10 15:01:30 -0600394 ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -0400395 if (ret)
396 return ret;
397
398 /* prepare the next free entry */
399 arena->freelist[lane].sub = 1 - arena->freelist[lane].sub;
400 if (++(arena->freelist[lane].seq) == 4)
401 arena->freelist[lane].seq = 1;
Vishal Vermad9b83c72017-08-30 19:36:03 -0600402 if (ent_e_flag(ent->old_map))
403 arena->freelist[lane].has_err = 1;
404 arena->freelist[lane].block = le32_to_cpu(ent_lba(ent->old_map));
Vishal Verma5212e112015-06-25 04:20:32 -0400405
406 return ret;
407}
408
409/*
410 * This function initializes the BTT map to the initial state, which is
411 * all-zeroes, and indicates an identity mapping
412 */
413static int btt_map_init(struct arena_info *arena)
414{
415 int ret = -EINVAL;
416 void *zerobuf;
417 size_t offset = 0;
418 size_t chunk_size = SZ_2M;
419 size_t mapsize = arena->logoff - arena->mapoff;
420
421 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
422 if (!zerobuf)
423 return -ENOMEM;
424
Vishal Vermab177fe82017-05-10 15:01:31 -0600425 /*
426 * mapoff should always be at least 512B aligned. We rely on that to
427 * make sure rw_bytes does error clearing correctly, so make sure that
428 * is the case.
429 */
Vishal Verma86652d22017-09-05 14:35:39 -0600430 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->mapoff, 512),
431 "arena->mapoff: %#llx is unaligned\n", arena->mapoff);
Vishal Vermab177fe82017-05-10 15:01:31 -0600432
Vishal Verma5212e112015-06-25 04:20:32 -0400433 while (mapsize) {
434 size_t size = min(mapsize, chunk_size);
435
Vishal Verma86652d22017-09-05 14:35:39 -0600436 dev_WARN_ONCE(to_dev(arena), size < 512,
Randy Dunlap04c3c982017-09-08 09:36:57 -0700437 "chunk size: %#zx is unaligned\n", size);
Vishal Verma5212e112015-06-25 04:20:32 -0400438 ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf,
Vishal Verma3ae3d672017-05-10 15:01:30 -0600439 size, 0);
Vishal Verma5212e112015-06-25 04:20:32 -0400440 if (ret)
441 goto free;
442
443 offset += size;
444 mapsize -= size;
445 cond_resched();
446 }
447
448 free:
449 kfree(zerobuf);
450 return ret;
451}
452
453/*
454 * This function initializes the BTT log with 'fake' entries pointing
455 * to the initial reserved set of blocks as being free
456 */
457static int btt_log_init(struct arena_info *arena)
458{
Vishal Vermab177fe82017-05-10 15:01:31 -0600459 size_t logsize = arena->info2off - arena->logoff;
460 size_t chunk_size = SZ_4K, offset = 0;
Vishal Verma166f39b2017-12-18 09:28:39 -0700461 struct log_entry ent;
Vishal Vermab177fe82017-05-10 15:01:31 -0600462 void *zerobuf;
Vishal Verma5212e112015-06-25 04:20:32 -0400463 int ret;
464 u32 i;
Vishal Verma5212e112015-06-25 04:20:32 -0400465
Vishal Vermab177fe82017-05-10 15:01:31 -0600466 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
467 if (!zerobuf)
468 return -ENOMEM;
469 /*
470 * logoff should always be at least 512B aligned. We rely on that to
471 * make sure rw_bytes does error clearing correctly, so make sure that
472 * is the case.
473 */
Vishal Verma86652d22017-09-05 14:35:39 -0600474 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->logoff, 512),
475 "arena->logoff: %#llx is unaligned\n", arena->logoff);
Vishal Vermab177fe82017-05-10 15:01:31 -0600476
477 while (logsize) {
478 size_t size = min(logsize, chunk_size);
479
Vishal Verma86652d22017-09-05 14:35:39 -0600480 dev_WARN_ONCE(to_dev(arena), size < 512,
Randy Dunlap04c3c982017-09-08 09:36:57 -0700481 "chunk size: %#zx is unaligned\n", size);
Vishal Vermab177fe82017-05-10 15:01:31 -0600482 ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf,
483 size, 0);
484 if (ret)
485 goto free;
486
487 offset += size;
488 logsize -= size;
489 cond_resched();
490 }
Vishal Verma5212e112015-06-25 04:20:32 -0400491
492 for (i = 0; i < arena->nfree; i++) {
Vishal Verma166f39b2017-12-18 09:28:39 -0700493 ent.lba = cpu_to_le32(i);
494 ent.old_map = cpu_to_le32(arena->external_nlba + i);
495 ent.new_map = cpu_to_le32(arena->external_nlba + i);
496 ent.seq = cpu_to_le32(LOG_SEQ_INIT);
497 ret = __btt_log_write(arena, i, 0, &ent, 0);
Vishal Verma5212e112015-06-25 04:20:32 -0400498 if (ret)
Vishal Vermab177fe82017-05-10 15:01:31 -0600499 goto free;
Vishal Verma5212e112015-06-25 04:20:32 -0400500 }
501
Vishal Vermab177fe82017-05-10 15:01:31 -0600502 free:
503 kfree(zerobuf);
504 return ret;
Vishal Verma5212e112015-06-25 04:20:32 -0400505}
506
Vishal Vermad9b83c72017-08-30 19:36:03 -0600507static u64 to_namespace_offset(struct arena_info *arena, u64 lba)
508{
509 return arena->dataoff + ((u64)lba * arena->internal_lbasize);
510}
511
512static int arena_clear_freelist_error(struct arena_info *arena, u32 lane)
513{
514 int ret = 0;
515
516 if (arena->freelist[lane].has_err) {
517 void *zero_page = page_address(ZERO_PAGE(0));
518 u32 lba = arena->freelist[lane].block;
519 u64 nsoff = to_namespace_offset(arena, lba);
520 unsigned long len = arena->sector_size;
521
522 mutex_lock(&arena->err_lock);
523
524 while (len) {
525 unsigned long chunk = min(len, PAGE_SIZE);
526
527 ret = arena_write_bytes(arena, nsoff, zero_page,
528 chunk, 0);
529 if (ret)
530 break;
531 len -= chunk;
532 nsoff += chunk;
533 if (len == 0)
534 arena->freelist[lane].has_err = 0;
535 }
536 mutex_unlock(&arena->err_lock);
537 }
538 return ret;
539}
540
Vishal Verma5212e112015-06-25 04:20:32 -0400541static int btt_freelist_init(struct arena_info *arena)
542{
543 int old, new, ret;
544 u32 i, map_entry;
545 struct log_entry log_new, log_old;
546
547 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
548 GFP_KERNEL);
549 if (!arena->freelist)
550 return -ENOMEM;
551
552 for (i = 0; i < arena->nfree; i++) {
553 old = btt_log_read(arena, i, &log_old, LOG_OLD_ENT);
554 if (old < 0)
555 return old;
556
557 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
558 if (new < 0)
559 return new;
560
561 /* sub points to the next one to be overwritten */
562 arena->freelist[i].sub = 1 - new;
563 arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq));
564 arena->freelist[i].block = le32_to_cpu(log_new.old_map);
565
Vishal Vermad9b83c72017-08-30 19:36:03 -0600566 /*
567 * FIXME: if error clearing fails during init, we want to make
568 * the BTT read-only
569 */
570 if (ent_e_flag(log_new.old_map)) {
571 ret = arena_clear_freelist_error(arena, i);
572 if (ret)
Vishal Verma86652d22017-09-05 14:35:39 -0600573 dev_err_ratelimited(to_dev(arena),
574 "Unable to clear known errors\n");
Vishal Vermad9b83c72017-08-30 19:36:03 -0600575 }
576
Vishal Verma5212e112015-06-25 04:20:32 -0400577 /* This implies a newly created or untouched flog entry */
578 if (log_new.old_map == log_new.new_map)
579 continue;
580
581 /* Check if map recovery is needed */
582 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
Vishal Verma3ae3d672017-05-10 15:01:30 -0600583 NULL, NULL, 0);
Vishal Verma5212e112015-06-25 04:20:32 -0400584 if (ret)
585 return ret;
586 if ((le32_to_cpu(log_new.new_map) != map_entry) &&
587 (le32_to_cpu(log_new.old_map) == map_entry)) {
588 /*
589 * Last transaction wrote the flog, but wasn't able
590 * to complete the map write. So fix up the map.
591 */
592 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
Vishal Verma3ae3d672017-05-10 15:01:30 -0600593 le32_to_cpu(log_new.new_map), 0, 0, 0);
Vishal Verma5212e112015-06-25 04:20:32 -0400594 if (ret)
595 return ret;
596 }
Vishal Verma5212e112015-06-25 04:20:32 -0400597 }
598
599 return 0;
600}
601
Vishal Verma166f39b2017-12-18 09:28:39 -0700602static bool ent_is_padding(struct log_entry *ent)
603{
604 return (ent->lba == 0) && (ent->old_map == 0) && (ent->new_map == 0)
605 && (ent->seq == 0);
606}
607
608/*
609 * Detecting valid log indices: We read a log group (see the comments in btt.h
610 * for a description of a 'log_group' and its 'slots'), and iterate over its
611 * four slots. We expect that a padding slot will be all-zeroes, and use this
612 * to detect a padding slot vs. an actual entry.
613 *
614 * If a log_group is in the initial state, i.e. hasn't been used since the
615 * creation of this BTT layout, it will have three of the four slots with
616 * zeroes. We skip over these log_groups for the detection of log_index. If
617 * all log_groups are in the initial state (i.e. the BTT has never been
618 * written to), it is safe to assume the 'new format' of log entries in slots
619 * (0, 1).
620 */
621static int log_set_indices(struct arena_info *arena)
622{
623 bool idx_set = false, initial_state = true;
624 int ret, log_index[2] = {-1, -1};
625 u32 i, j, next_idx = 0;
626 struct log_group log;
627 u32 pad_count = 0;
628
629 for (i = 0; i < arena->nfree; i++) {
630 ret = btt_log_group_read(arena, i, &log);
631 if (ret < 0)
632 return ret;
633
634 for (j = 0; j < 4; j++) {
635 if (!idx_set) {
636 if (ent_is_padding(&log.ent[j])) {
637 pad_count++;
638 continue;
639 } else {
640 /* Skip if index has been recorded */
641 if ((next_idx == 1) &&
642 (j == log_index[0]))
643 continue;
644 /* valid entry, record index */
645 log_index[next_idx] = j;
646 next_idx++;
647 }
648 if (next_idx == 2) {
649 /* two valid entries found */
650 idx_set = true;
651 } else if (next_idx > 2) {
652 /* too many valid indices */
653 return -ENXIO;
654 }
655 } else {
656 /*
657 * once the indices have been set, just verify
658 * that all subsequent log groups are either in
659 * their initial state or follow the same
660 * indices.
661 */
662 if (j == log_index[0]) {
663 /* entry must be 'valid' */
664 if (ent_is_padding(&log.ent[j]))
665 return -ENXIO;
666 } else if (j == log_index[1]) {
667 ;
668 /*
669 * log_index[1] can be padding if the
670 * lane never got used and it is still
671 * in the initial state (three 'padding'
672 * entries)
673 */
674 } else {
675 /* entry must be invalid (padding) */
676 if (!ent_is_padding(&log.ent[j]))
677 return -ENXIO;
678 }
679 }
680 }
681 /*
682 * If any of the log_groups have more than one valid,
683 * non-padding entry, then the we are no longer in the
684 * initial_state
685 */
686 if (pad_count < 3)
687 initial_state = false;
688 pad_count = 0;
689 }
690
691 if (!initial_state && !idx_set)
692 return -ENXIO;
693
694 /*
695 * If all the entries in the log were in the initial state,
696 * assume new padding scheme
697 */
698 if (initial_state)
699 log_index[1] = 1;
700
701 /*
702 * Only allow the known permutations of log/padding indices,
703 * i.e. (0, 1), and (0, 2)
704 */
705 if ((log_index[0] == 0) && ((log_index[1] == 1) || (log_index[1] == 2)))
706 ; /* known index possibilities */
707 else {
708 dev_err(to_dev(arena), "Found an unknown padding scheme\n");
709 return -ENXIO;
710 }
711
712 arena->log_index[0] = log_index[0];
713 arena->log_index[1] = log_index[1];
714 dev_dbg(to_dev(arena), "log_index_0 = %d\n", log_index[0]);
715 dev_dbg(to_dev(arena), "log_index_1 = %d\n", log_index[1]);
716 return 0;
717}
718
Vishal Verma5212e112015-06-25 04:20:32 -0400719static int btt_rtt_init(struct arena_info *arena)
720{
721 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
722 if (arena->rtt == NULL)
723 return -ENOMEM;
724
725 return 0;
726}
727
728static int btt_maplocks_init(struct arena_info *arena)
729{
730 u32 i;
731
732 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
733 GFP_KERNEL);
734 if (!arena->map_locks)
735 return -ENOMEM;
736
737 for (i = 0; i < arena->nfree; i++)
738 spin_lock_init(&arena->map_locks[i].lock);
739
740 return 0;
741}
742
743static struct arena_info *alloc_arena(struct btt *btt, size_t size,
744 size_t start, size_t arena_off)
745{
746 struct arena_info *arena;
747 u64 logsize, mapsize, datasize;
748 u64 available = size;
749
750 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
751 if (!arena)
752 return NULL;
753 arena->nd_btt = btt->nd_btt;
Vishal Verma75892002017-08-30 19:36:01 -0600754 arena->sector_size = btt->sector_size;
Vishal Verma5212e112015-06-25 04:20:32 -0400755
756 if (!size)
757 return arena;
758
759 arena->size = size;
760 arena->external_lba_start = start;
761 arena->external_lbasize = btt->lbasize;
762 arena->internal_lbasize = roundup(arena->external_lbasize,
763 INT_LBASIZE_ALIGNMENT);
764 arena->nfree = BTT_DEFAULT_NFREE;
Vishal Verma14e49452017-06-28 14:25:00 -0600765 arena->version_major = btt->nd_btt->version_major;
766 arena->version_minor = btt->nd_btt->version_minor;
Vishal Verma5212e112015-06-25 04:20:32 -0400767
768 if (available % BTT_PG_SIZE)
769 available -= (available % BTT_PG_SIZE);
770
771 /* Two pages are reserved for the super block and its copy */
772 available -= 2 * BTT_PG_SIZE;
773
774 /* The log takes a fixed amount of space based on nfree */
Vishal Verma166f39b2017-12-18 09:28:39 -0700775 logsize = roundup(arena->nfree * LOG_GRP_SIZE, BTT_PG_SIZE);
Vishal Verma5212e112015-06-25 04:20:32 -0400776 available -= logsize;
777
778 /* Calculate optimal split between map and data area */
779 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
780 arena->internal_lbasize + MAP_ENT_SIZE);
781 arena->external_nlba = arena->internal_nlba - arena->nfree;
782
783 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
784 datasize = available - mapsize;
785
786 /* 'Absolute' values, relative to start of storage space */
787 arena->infooff = arena_off;
788 arena->dataoff = arena->infooff + BTT_PG_SIZE;
789 arena->mapoff = arena->dataoff + datasize;
790 arena->logoff = arena->mapoff + mapsize;
791 arena->info2off = arena->logoff + logsize;
Vishal Verma166f39b2017-12-18 09:28:39 -0700792
793 /* Default log indices are (0,1) */
794 arena->log_index[0] = 0;
795 arena->log_index[1] = 1;
Vishal Verma5212e112015-06-25 04:20:32 -0400796 return arena;
797}
798
799static void free_arenas(struct btt *btt)
800{
801 struct arena_info *arena, *next;
802
803 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
804 list_del(&arena->list);
805 kfree(arena->rtt);
806 kfree(arena->map_locks);
807 kfree(arena->freelist);
808 debugfs_remove_recursive(arena->debugfs_dir);
809 kfree(arena);
810 }
811}
812
813/*
Vishal Verma5212e112015-06-25 04:20:32 -0400814 * This function reads an existing valid btt superblock and
815 * populates the corresponding arena_info struct
816 */
817static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
818 u64 arena_off)
819{
820 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
821 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
822 arena->external_nlba = le32_to_cpu(super->external_nlba);
823 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
824 arena->nfree = le32_to_cpu(super->nfree);
825 arena->version_major = le16_to_cpu(super->version_major);
826 arena->version_minor = le16_to_cpu(super->version_minor);
827
828 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
829 le64_to_cpu(super->nextoff));
830 arena->infooff = arena_off;
831 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
832 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
833 arena->logoff = arena_off + le64_to_cpu(super->logoff);
834 arena->info2off = arena_off + le64_to_cpu(super->info2off);
835
Dan Williams5e329402015-07-11 10:02:46 -0400836 arena->size = (le64_to_cpu(super->nextoff) > 0)
837 ? (le64_to_cpu(super->nextoff))
838 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
Vishal Verma5212e112015-06-25 04:20:32 -0400839
840 arena->flags = le32_to_cpu(super->flags);
841}
842
843static int discover_arenas(struct btt *btt)
844{
845 int ret = 0;
846 struct arena_info *arena;
847 struct btt_sb *super;
848 size_t remaining = btt->rawsize;
849 u64 cur_nlba = 0;
850 size_t cur_off = 0;
851 int num_arenas = 0;
852
853 super = kzalloc(sizeof(*super), GFP_KERNEL);
854 if (!super)
855 return -ENOMEM;
856
857 while (remaining) {
858 /* Alloc memory for arena */
859 arena = alloc_arena(btt, 0, 0, 0);
860 if (!arena) {
861 ret = -ENOMEM;
862 goto out_super;
863 }
864
865 arena->infooff = cur_off;
866 ret = btt_info_read(arena, super);
867 if (ret)
868 goto out;
869
Vishal Vermaab45e762015-07-29 14:58:08 -0600870 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
Vishal Verma5212e112015-06-25 04:20:32 -0400871 if (remaining == btt->rawsize) {
872 btt->init_state = INIT_NOTFOUND;
873 dev_info(to_dev(arena), "No existing arenas\n");
874 goto out;
875 } else {
Vishal Vermae6be2dc2017-06-30 18:32:51 -0600876 dev_err(to_dev(arena),
Vishal Verma5212e112015-06-25 04:20:32 -0400877 "Found corrupted metadata!\n");
878 ret = -ENODEV;
879 goto out;
880 }
881 }
882
883 arena->external_lba_start = cur_nlba;
884 parse_arena_meta(arena, super, cur_off);
885
Vishal Verma166f39b2017-12-18 09:28:39 -0700886 ret = log_set_indices(arena);
887 if (ret) {
888 dev_err(to_dev(arena),
889 "Unable to deduce log/padding indices\n");
890 goto out;
891 }
892
Vishal Vermad9b83c72017-08-30 19:36:03 -0600893 mutex_init(&arena->err_lock);
Vishal Verma5212e112015-06-25 04:20:32 -0400894 ret = btt_freelist_init(arena);
895 if (ret)
896 goto out;
897
898 ret = btt_rtt_init(arena);
899 if (ret)
900 goto out;
901
902 ret = btt_maplocks_init(arena);
903 if (ret)
904 goto out;
905
906 list_add_tail(&arena->list, &btt->arena_list);
907
908 remaining -= arena->size;
909 cur_off += arena->size;
910 cur_nlba += arena->external_nlba;
911 num_arenas++;
912
913 if (arena->nextoff == 0)
914 break;
915 }
916 btt->num_arenas = num_arenas;
917 btt->nlba = cur_nlba;
918 btt->init_state = INIT_READY;
919
920 kfree(super);
921 return ret;
922
923 out:
924 kfree(arena);
925 free_arenas(btt);
926 out_super:
927 kfree(super);
928 return ret;
929}
930
931static int create_arenas(struct btt *btt)
932{
933 size_t remaining = btt->rawsize;
934 size_t cur_off = 0;
935
936 while (remaining) {
937 struct arena_info *arena;
938 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
939
940 remaining -= arena_size;
941 if (arena_size < ARENA_MIN_SIZE)
942 break;
943
944 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
945 if (!arena) {
946 free_arenas(btt);
947 return -ENOMEM;
948 }
949 btt->nlba += arena->external_nlba;
950 if (remaining >= ARENA_MIN_SIZE)
951 arena->nextoff = arena->size;
952 else
953 arena->nextoff = 0;
954 cur_off += arena_size;
955 list_add_tail(&arena->list, &btt->arena_list);
956 }
957
958 return 0;
959}
960
961/*
962 * This function completes arena initialization by writing
963 * all the metadata.
964 * It is only called for an uninitialized arena when a write
965 * to that arena occurs for the first time.
966 */
Vishal Vermafbde1412015-07-29 14:58:07 -0600967static int btt_arena_write_layout(struct arena_info *arena)
Vishal Verma5212e112015-06-25 04:20:32 -0400968{
969 int ret;
Dan Williamse1455742015-07-30 17:57:47 -0400970 u64 sum;
Vishal Verma5212e112015-06-25 04:20:32 -0400971 struct btt_sb *super;
Vishal Vermafbde1412015-07-29 14:58:07 -0600972 struct nd_btt *nd_btt = arena->nd_btt;
Vishal Verma6ec68952015-07-29 14:58:09 -0600973 const u8 *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
Vishal Verma5212e112015-06-25 04:20:32 -0400974
975 ret = btt_map_init(arena);
976 if (ret)
977 return ret;
978
979 ret = btt_log_init(arena);
980 if (ret)
981 return ret;
982
983 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
984 if (!super)
985 return -ENOMEM;
986
987 strncpy(super->signature, BTT_SIG, BTT_SIG_LEN);
Vishal Vermafbde1412015-07-29 14:58:07 -0600988 memcpy(super->uuid, nd_btt->uuid, 16);
Vishal Verma6ec68952015-07-29 14:58:09 -0600989 memcpy(super->parent_uuid, parent_uuid, 16);
Vishal Verma5212e112015-06-25 04:20:32 -0400990 super->flags = cpu_to_le32(arena->flags);
991 super->version_major = cpu_to_le16(arena->version_major);
992 super->version_minor = cpu_to_le16(arena->version_minor);
993 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
994 super->external_nlba = cpu_to_le32(arena->external_nlba);
995 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
996 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
997 super->nfree = cpu_to_le32(arena->nfree);
998 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
999 super->nextoff = cpu_to_le64(arena->nextoff);
1000 /*
1001 * Subtract arena->infooff (arena start) so numbers are relative
1002 * to 'this' arena
1003 */
1004 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
1005 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
1006 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
1007 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
1008
1009 super->flags = 0;
Dan Williamse1455742015-07-30 17:57:47 -04001010 sum = nd_sb_checksum((struct nd_gen_sb *) super);
1011 super->checksum = cpu_to_le64(sum);
Vishal Verma5212e112015-06-25 04:20:32 -04001012
1013 ret = btt_info_write(arena, super);
1014
1015 kfree(super);
1016 return ret;
1017}
1018
1019/*
1020 * This function completes the initialization for the BTT namespace
1021 * such that it is ready to accept IOs
1022 */
1023static int btt_meta_init(struct btt *btt)
1024{
1025 int ret = 0;
1026 struct arena_info *arena;
1027
1028 mutex_lock(&btt->init_lock);
1029 list_for_each_entry(arena, &btt->arena_list, list) {
Vishal Vermafbde1412015-07-29 14:58:07 -06001030 ret = btt_arena_write_layout(arena);
Vishal Verma5212e112015-06-25 04:20:32 -04001031 if (ret)
1032 goto unlock;
1033
1034 ret = btt_freelist_init(arena);
1035 if (ret)
1036 goto unlock;
1037
1038 ret = btt_rtt_init(arena);
1039 if (ret)
1040 goto unlock;
1041
1042 ret = btt_maplocks_init(arena);
1043 if (ret)
1044 goto unlock;
1045 }
1046
1047 btt->init_state = INIT_READY;
1048
1049 unlock:
1050 mutex_unlock(&btt->init_lock);
1051 return ret;
1052}
1053
Vishal Verma41cd8b72015-06-25 04:21:52 -04001054static u32 btt_meta_size(struct btt *btt)
1055{
1056 return btt->lbasize - btt->sector_size;
1057}
1058
Vishal Verma5212e112015-06-25 04:20:32 -04001059/*
1060 * This function calculates the arena in which the given LBA lies
1061 * by doing a linear walk. This is acceptable since we expect only
1062 * a few arenas. If we have backing devices that get much larger,
1063 * we can construct a balanced binary tree of arenas at init time
1064 * so that this range search becomes faster.
1065 */
1066static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
1067 struct arena_info **arena)
1068{
1069 struct arena_info *arena_list;
1070 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
1071
1072 list_for_each_entry(arena_list, &btt->arena_list, list) {
1073 if (lba < arena_list->external_nlba) {
1074 *arena = arena_list;
1075 *premap = lba;
1076 return 0;
1077 }
1078 lba -= arena_list->external_nlba;
1079 }
1080
1081 return -EIO;
1082}
1083
1084/*
1085 * The following (lock_map, unlock_map) are mostly just to improve
1086 * readability, since they index into an array of locks
1087 */
1088static void lock_map(struct arena_info *arena, u32 premap)
1089 __acquires(&arena->map_locks[idx].lock)
1090{
1091 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1092
1093 spin_lock(&arena->map_locks[idx].lock);
1094}
1095
1096static void unlock_map(struct arena_info *arena, u32 premap)
1097 __releases(&arena->map_locks[idx].lock)
1098{
1099 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1100
1101 spin_unlock(&arena->map_locks[idx].lock);
1102}
1103
Vishal Verma5212e112015-06-25 04:20:32 -04001104static int btt_data_read(struct arena_info *arena, struct page *page,
1105 unsigned int off, u32 lba, u32 len)
1106{
1107 int ret;
1108 u64 nsoff = to_namespace_offset(arena, lba);
1109 void *mem = kmap_atomic(page);
1110
Vishal Verma3ae3d672017-05-10 15:01:30 -06001111 ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -04001112 kunmap_atomic(mem);
1113
1114 return ret;
1115}
1116
1117static int btt_data_write(struct arena_info *arena, u32 lba,
1118 struct page *page, unsigned int off, u32 len)
1119{
1120 int ret;
1121 u64 nsoff = to_namespace_offset(arena, lba);
1122 void *mem = kmap_atomic(page);
1123
Vishal Verma3ae3d672017-05-10 15:01:30 -06001124 ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -04001125 kunmap_atomic(mem);
1126
1127 return ret;
1128}
1129
1130static void zero_fill_data(struct page *page, unsigned int off, u32 len)
1131{
1132 void *mem = kmap_atomic(page);
1133
1134 memset(mem + off, 0, len);
1135 kunmap_atomic(mem);
1136}
1137
Vishal Verma41cd8b72015-06-25 04:21:52 -04001138#ifdef CONFIG_BLK_DEV_INTEGRITY
1139static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1140 struct arena_info *arena, u32 postmap, int rw)
1141{
1142 unsigned int len = btt_meta_size(btt);
1143 u64 meta_nsoff;
1144 int ret = 0;
1145
1146 if (bip == NULL)
1147 return 0;
1148
1149 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
1150
1151 while (len) {
1152 unsigned int cur_len;
1153 struct bio_vec bv;
1154 void *mem;
1155
1156 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1157 /*
1158 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
1159 * .bv_offset already adjusted for iter->bi_bvec_done, and we
1160 * can use those directly
1161 */
1162
1163 cur_len = min(len, bv.bv_len);
1164 mem = kmap_atomic(bv.bv_page);
1165 if (rw)
1166 ret = arena_write_bytes(arena, meta_nsoff,
Vishal Verma3ae3d672017-05-10 15:01:30 -06001167 mem + bv.bv_offset, cur_len,
1168 NVDIMM_IO_ATOMIC);
Vishal Verma41cd8b72015-06-25 04:21:52 -04001169 else
1170 ret = arena_read_bytes(arena, meta_nsoff,
Vishal Verma3ae3d672017-05-10 15:01:30 -06001171 mem + bv.bv_offset, cur_len,
1172 NVDIMM_IO_ATOMIC);
Vishal Verma41cd8b72015-06-25 04:21:52 -04001173
1174 kunmap_atomic(mem);
1175 if (ret)
1176 return ret;
1177
1178 len -= cur_len;
1179 meta_nsoff += cur_len;
Dmitry Monakhovb1fb2c52017-06-29 11:31:13 -07001180 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
1181 return -EIO;
Vishal Verma41cd8b72015-06-25 04:21:52 -04001182 }
1183
1184 return ret;
1185}
1186
1187#else /* CONFIG_BLK_DEV_INTEGRITY */
1188static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1189 struct arena_info *arena, u32 postmap, int rw)
1190{
1191 return 0;
1192}
1193#endif
1194
1195static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
1196 struct page *page, unsigned int off, sector_t sector,
1197 unsigned int len)
Vishal Verma5212e112015-06-25 04:20:32 -04001198{
1199 int ret = 0;
1200 int t_flag, e_flag;
1201 struct arena_info *arena = NULL;
1202 u32 lane = 0, premap, postmap;
1203
1204 while (len) {
1205 u32 cur_len;
1206
1207 lane = nd_region_acquire_lane(btt->nd_region);
1208
1209 ret = lba_to_arena(btt, sector, &premap, &arena);
1210 if (ret)
1211 goto out_lane;
1212
1213 cur_len = min(btt->sector_size, len);
1214
Vishal Verma3ae3d672017-05-10 15:01:30 -06001215 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag,
1216 NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -04001217 if (ret)
1218 goto out_lane;
1219
1220 /*
1221 * We loop to make sure that the post map LBA didn't change
1222 * from under us between writing the RTT and doing the actual
1223 * read.
1224 */
1225 while (1) {
1226 u32 new_map;
Vishal Verma13981992017-08-30 19:36:00 -06001227 int new_t, new_e;
Vishal Verma5212e112015-06-25 04:20:32 -04001228
1229 if (t_flag) {
1230 zero_fill_data(page, off, cur_len);
1231 goto out_lane;
1232 }
1233
1234 if (e_flag) {
1235 ret = -EIO;
1236 goto out_lane;
1237 }
1238
1239 arena->rtt[lane] = RTT_VALID | postmap;
1240 /*
1241 * Barrier to make sure this write is not reordered
1242 * to do the verification map_read before the RTT store
1243 */
1244 barrier();
1245
Vishal Verma13981992017-08-30 19:36:00 -06001246 ret = btt_map_read(arena, premap, &new_map, &new_t,
1247 &new_e, NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -04001248 if (ret)
1249 goto out_rtt;
1250
Vishal Verma13981992017-08-30 19:36:00 -06001251 if ((postmap == new_map) && (t_flag == new_t) &&
1252 (e_flag == new_e))
Vishal Verma5212e112015-06-25 04:20:32 -04001253 break;
1254
1255 postmap = new_map;
Vishal Verma13981992017-08-30 19:36:00 -06001256 t_flag = new_t;
1257 e_flag = new_e;
Vishal Verma5212e112015-06-25 04:20:32 -04001258 }
1259
1260 ret = btt_data_read(arena, page, off, postmap, cur_len);
Vishal Vermad9b83c72017-08-30 19:36:03 -06001261 if (ret) {
1262 int rc;
1263
1264 /* Media error - set the e_flag */
1265 rc = btt_map_write(arena, premap, postmap, 0, 1,
1266 NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -04001267 goto out_rtt;
Vishal Vermad9b83c72017-08-30 19:36:03 -06001268 }
Vishal Verma5212e112015-06-25 04:20:32 -04001269
Vishal Verma41cd8b72015-06-25 04:21:52 -04001270 if (bip) {
1271 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1272 if (ret)
1273 goto out_rtt;
1274 }
1275
Vishal Verma5212e112015-06-25 04:20:32 -04001276 arena->rtt[lane] = RTT_INVALID;
1277 nd_region_release_lane(btt->nd_region, lane);
1278
1279 len -= cur_len;
1280 off += cur_len;
1281 sector += btt->sector_size >> SECTOR_SHIFT;
1282 }
1283
1284 return 0;
1285
1286 out_rtt:
1287 arena->rtt[lane] = RTT_INVALID;
1288 out_lane:
1289 nd_region_release_lane(btt->nd_region, lane);
1290 return ret;
1291}
1292
Vishal Vermad9b83c72017-08-30 19:36:03 -06001293/*
1294 * Normally, arena_{read,write}_bytes will take care of the initial offset
1295 * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem,
1296 * we need the final, raw namespace offset here
1297 */
1298static bool btt_is_badblock(struct btt *btt, struct arena_info *arena,
1299 u32 postmap)
1300{
1301 u64 nsoff = adjust_initial_offset(arena->nd_btt,
1302 to_namespace_offset(arena, postmap));
1303 sector_t phys_sector = nsoff >> 9;
1304
1305 return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize);
1306}
1307
Vishal Verma41cd8b72015-06-25 04:21:52 -04001308static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1309 sector_t sector, struct page *page, unsigned int off,
1310 unsigned int len)
Vishal Verma5212e112015-06-25 04:20:32 -04001311{
1312 int ret = 0;
1313 struct arena_info *arena = NULL;
1314 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1315 struct log_entry log;
1316 int sub;
1317
1318 while (len) {
1319 u32 cur_len;
Vishal Vermad9b83c72017-08-30 19:36:03 -06001320 int e_flag;
Vishal Verma5212e112015-06-25 04:20:32 -04001321
Vishal Vermad9b83c72017-08-30 19:36:03 -06001322 retry:
Vishal Verma5212e112015-06-25 04:20:32 -04001323 lane = nd_region_acquire_lane(btt->nd_region);
1324
1325 ret = lba_to_arena(btt, sector, &premap, &arena);
1326 if (ret)
1327 goto out_lane;
1328 cur_len = min(btt->sector_size, len);
1329
1330 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1331 ret = -EIO;
1332 goto out_lane;
1333 }
1334
Vishal Vermad9b83c72017-08-30 19:36:03 -06001335 if (btt_is_badblock(btt, arena, arena->freelist[lane].block))
1336 arena->freelist[lane].has_err = 1;
1337
1338 if (mutex_is_locked(&arena->err_lock)
1339 || arena->freelist[lane].has_err) {
1340 nd_region_release_lane(btt->nd_region, lane);
1341
1342 ret = arena_clear_freelist_error(arena, lane);
1343 if (ret)
1344 return ret;
1345
1346 /* OK to acquire a different lane/free block */
1347 goto retry;
1348 }
1349
Vishal Verma5212e112015-06-25 04:20:32 -04001350 new_postmap = arena->freelist[lane].block;
1351
1352 /* Wait if the new block is being read from */
1353 for (i = 0; i < arena->nfree; i++)
1354 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1355 cpu_relax();
1356
1357
1358 if (new_postmap >= arena->internal_nlba) {
1359 ret = -EIO;
1360 goto out_lane;
Vishal Verma41cd8b72015-06-25 04:21:52 -04001361 }
1362
1363 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
Vishal Verma5212e112015-06-25 04:20:32 -04001364 if (ret)
1365 goto out_lane;
1366
Vishal Verma41cd8b72015-06-25 04:21:52 -04001367 if (bip) {
1368 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1369 WRITE);
1370 if (ret)
1371 goto out_lane;
1372 }
1373
Vishal Verma5212e112015-06-25 04:20:32 -04001374 lock_map(arena, premap);
Vishal Vermad9b83c72017-08-30 19:36:03 -06001375 ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag,
Vishal Verma3ae3d672017-05-10 15:01:30 -06001376 NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -04001377 if (ret)
1378 goto out_map;
1379 if (old_postmap >= arena->internal_nlba) {
1380 ret = -EIO;
1381 goto out_map;
1382 }
Vishal Vermad9b83c72017-08-30 19:36:03 -06001383 if (e_flag)
1384 set_e_flag(old_postmap);
Vishal Verma5212e112015-06-25 04:20:32 -04001385
1386 log.lba = cpu_to_le32(premap);
1387 log.old_map = cpu_to_le32(old_postmap);
1388 log.new_map = cpu_to_le32(new_postmap);
1389 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1390 sub = arena->freelist[lane].sub;
1391 ret = btt_flog_write(arena, lane, sub, &log);
1392 if (ret)
1393 goto out_map;
1394
Vishal Verma1db1f3c2017-08-30 19:35:58 -06001395 ret = btt_map_write(arena, premap, new_postmap, 0, 0,
1396 NVDIMM_IO_ATOMIC);
Vishal Verma5212e112015-06-25 04:20:32 -04001397 if (ret)
1398 goto out_map;
1399
1400 unlock_map(arena, premap);
1401 nd_region_release_lane(btt->nd_region, lane);
1402
Vishal Vermad9b83c72017-08-30 19:36:03 -06001403 if (e_flag) {
1404 ret = arena_clear_freelist_error(arena, lane);
1405 if (ret)
1406 return ret;
1407 }
1408
Vishal Verma5212e112015-06-25 04:20:32 -04001409 len -= cur_len;
1410 off += cur_len;
1411 sector += btt->sector_size >> SECTOR_SHIFT;
1412 }
1413
1414 return 0;
1415
1416 out_map:
1417 unlock_map(arena, premap);
1418 out_lane:
1419 nd_region_release_lane(btt->nd_region, lane);
1420 return ret;
1421}
1422
Vishal Verma41cd8b72015-06-25 04:21:52 -04001423static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1424 struct page *page, unsigned int len, unsigned int off,
Jens Axboec11f0c02016-08-05 08:11:04 -06001425 bool is_write, sector_t sector)
Vishal Verma5212e112015-06-25 04:20:32 -04001426{
1427 int ret;
1428
Jens Axboec11f0c02016-08-05 08:11:04 -06001429 if (!is_write) {
Vishal Verma41cd8b72015-06-25 04:21:52 -04001430 ret = btt_read_pg(btt, bip, page, off, sector, len);
Vishal Verma5212e112015-06-25 04:20:32 -04001431 flush_dcache_page(page);
1432 } else {
1433 flush_dcache_page(page);
Vishal Verma41cd8b72015-06-25 04:21:52 -04001434 ret = btt_write_pg(btt, bip, sector, page, off, len);
Vishal Verma5212e112015-06-25 04:20:32 -04001435 }
1436
1437 return ret;
1438}
1439
Jens Axboedece1632015-11-05 10:41:16 -07001440static blk_qc_t btt_make_request(struct request_queue *q, struct bio *bio)
Vishal Verma5212e112015-06-25 04:20:32 -04001441{
Vishal Verma41cd8b72015-06-25 04:21:52 -04001442 struct bio_integrity_payload *bip = bio_integrity(bio);
Vishal Verma5212e112015-06-25 04:20:32 -04001443 struct btt *btt = q->queuedata;
1444 struct bvec_iter iter;
Dan Williamsf0dc0892015-05-16 12:28:53 -04001445 unsigned long start;
Vishal Verma5212e112015-06-25 04:20:32 -04001446 struct bio_vec bvec;
Mike Christieabf54542016-08-04 14:23:34 -06001447 int err = 0;
Dan Williamsf0dc0892015-05-16 12:28:53 -04001448 bool do_acct;
Vishal Verma5212e112015-06-25 04:20:32 -04001449
Dmitry Monakhove23947b2017-06-29 11:31:11 -07001450 if (!bio_integrity_prep(bio))
1451 return BLK_QC_T_NONE;
Vishal Verma41cd8b72015-06-25 04:21:52 -04001452
Dan Williamsf0dc0892015-05-16 12:28:53 -04001453 do_acct = nd_iostat_start(bio, &start);
Vishal Verma5212e112015-06-25 04:20:32 -04001454 bio_for_each_segment(bvec, bio, iter) {
1455 unsigned int len = bvec.bv_len;
1456
Vishal Verma86652d22017-09-05 14:35:39 -06001457 if (len > PAGE_SIZE || len < btt->sector_size ||
1458 len % btt->sector_size) {
1459 dev_err_ratelimited(&btt->nd_btt->dev,
1460 "unaligned bio segment (len: %d)\n", len);
1461 bio->bi_status = BLK_STS_IOERR;
1462 break;
1463 }
Vishal Verma5212e112015-06-25 04:20:32 -04001464
Vishal Verma41cd8b72015-06-25 04:21:52 -04001465 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
Jens Axboec11f0c02016-08-05 08:11:04 -06001466 op_is_write(bio_op(bio)), iter.bi_sector);
Vishal Verma5212e112015-06-25 04:20:32 -04001467 if (err) {
Vishal Vermae6be2dc2017-06-30 18:32:51 -06001468 dev_err(&btt->nd_btt->dev,
Vishal Verma5212e112015-06-25 04:20:32 -04001469 "io error in %s sector %lld, len %d,\n",
Mike Christieabf54542016-08-04 14:23:34 -06001470 (op_is_write(bio_op(bio))) ? "WRITE" :
1471 "READ",
Vishal Verma5212e112015-06-25 04:20:32 -04001472 (unsigned long long) iter.bi_sector, len);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02001473 bio->bi_status = errno_to_blk_status(err);
Dan Williamsf0dc0892015-05-16 12:28:53 -04001474 break;
Vishal Verma5212e112015-06-25 04:20:32 -04001475 }
1476 }
Dan Williamsf0dc0892015-05-16 12:28:53 -04001477 if (do_acct)
1478 nd_iostat_end(bio, start);
Vishal Verma5212e112015-06-25 04:20:32 -04001479
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02001480 bio_endio(bio);
Jens Axboedece1632015-11-05 10:41:16 -07001481 return BLK_QC_T_NONE;
Vishal Verma5212e112015-06-25 04:20:32 -04001482}
1483
1484static int btt_rw_page(struct block_device *bdev, sector_t sector,
Jens Axboec11f0c02016-08-05 08:11:04 -06001485 struct page *page, bool is_write)
Vishal Verma5212e112015-06-25 04:20:32 -04001486{
1487 struct btt *btt = bdev->bd_disk->private_data;
Vishal Vermac13c43d2017-06-29 16:59:11 -06001488 int rc;
Huang Ying98cc0932017-09-06 16:22:27 -07001489 unsigned int len;
Vishal Verma5212e112015-06-25 04:20:32 -04001490
Huang Ying98cc0932017-09-06 16:22:27 -07001491 len = hpage_nr_pages(page) * PAGE_SIZE;
1492 rc = btt_do_bvec(btt, NULL, page, len, 0, is_write, sector);
Vishal Vermac13c43d2017-06-29 16:59:11 -06001493 if (rc == 0)
1494 page_endio(page, is_write, 0);
1495
1496 return rc;
Vishal Verma5212e112015-06-25 04:20:32 -04001497}
1498
1499
1500static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1501{
1502 /* some standard values */
1503 geo->heads = 1 << 6;
1504 geo->sectors = 1 << 5;
1505 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1506 return 0;
1507}
1508
1509static const struct block_device_operations btt_fops = {
1510 .owner = THIS_MODULE,
1511 .rw_page = btt_rw_page,
1512 .getgeo = btt_getgeo,
Dan Williams58138822015-06-23 20:08:34 -04001513 .revalidate_disk = nvdimm_revalidate_disk,
Vishal Verma5212e112015-06-25 04:20:32 -04001514};
1515
1516static int btt_blk_init(struct btt *btt)
1517{
1518 struct nd_btt *nd_btt = btt->nd_btt;
1519 struct nd_namespace_common *ndns = nd_btt->ndns;
1520
1521 /* create a new disk and request queue for btt */
1522 btt->btt_queue = blk_alloc_queue(GFP_KERNEL);
1523 if (!btt->btt_queue)
1524 return -ENOMEM;
1525
1526 btt->btt_disk = alloc_disk(0);
1527 if (!btt->btt_disk) {
1528 blk_cleanup_queue(btt->btt_queue);
1529 return -ENOMEM;
1530 }
1531
1532 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
Vishal Verma5212e112015-06-25 04:20:32 -04001533 btt->btt_disk->first_minor = 0;
1534 btt->btt_disk->fops = &btt_fops;
1535 btt->btt_disk->private_data = btt;
1536 btt->btt_disk->queue = btt->btt_queue;
1537 btt->btt_disk->flags = GENHD_FL_EXT_DEVT;
1538
1539 blk_queue_make_request(btt->btt_queue, btt_make_request);
1540 blk_queue_logical_block_size(btt->btt_queue, btt->sector_size);
1541 blk_queue_max_hw_sectors(btt->btt_queue, UINT_MAX);
Vishal Verma5212e112015-06-25 04:20:32 -04001542 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, btt->btt_queue);
1543 btt->btt_queue->queuedata = btt;
1544
Vishal Verma41cd8b72015-06-25 04:21:52 -04001545 if (btt_meta_size(btt)) {
1546 int rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1547
1548 if (rc) {
1549 del_gendisk(btt->btt_disk);
1550 put_disk(btt->btt_disk);
1551 blk_cleanup_queue(btt->btt_queue);
1552 return rc;
1553 }
1554 }
1555 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
Vishal Verma865a0222018-03-05 16:56:13 -07001556 device_add_disk(&btt->nd_btt->dev, btt->btt_disk);
Vishal Vermaabe8b4e2016-07-27 16:38:59 -06001557 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
Dan Williams58138822015-06-23 20:08:34 -04001558 revalidate_disk(btt->btt_disk);
Vishal Verma5212e112015-06-25 04:20:32 -04001559
1560 return 0;
1561}
1562
1563static void btt_blk_cleanup(struct btt *btt)
1564{
Vishal Verma5212e112015-06-25 04:20:32 -04001565 del_gendisk(btt->btt_disk);
1566 put_disk(btt->btt_disk);
1567 blk_cleanup_queue(btt->btt_queue);
1568}
1569
1570/**
1571 * btt_init - initialize a block translation table for the given device
1572 * @nd_btt: device with BTT geometry and backing device info
1573 * @rawsize: raw size in bytes of the backing device
1574 * @lbasize: lba size of the backing device
1575 * @uuid: A uuid for the backing device - this is stored on media
1576 * @maxlane: maximum number of parallel requests the device can handle
1577 *
1578 * Initialize a Block Translation Table on a backing device to provide
1579 * single sector power fail atomicity.
1580 *
1581 * Context:
1582 * Might sleep.
1583 *
1584 * Returns:
1585 * Pointer to a new struct btt on success, NULL on failure.
1586 */
1587static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1588 u32 lbasize, u8 *uuid, struct nd_region *nd_region)
1589{
1590 int ret;
1591 struct btt *btt;
Vishal Vermad9b83c72017-08-30 19:36:03 -06001592 struct nd_namespace_io *nsio;
Vishal Verma5212e112015-06-25 04:20:32 -04001593 struct device *dev = &nd_btt->dev;
1594
Dan Williamse32bc722016-03-17 18:23:09 -07001595 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
Vishal Verma5212e112015-06-25 04:20:32 -04001596 if (!btt)
1597 return NULL;
1598
1599 btt->nd_btt = nd_btt;
1600 btt->rawsize = rawsize;
1601 btt->lbasize = lbasize;
1602 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1603 INIT_LIST_HEAD(&btt->arena_list);
1604 mutex_init(&btt->init_lock);
1605 btt->nd_region = nd_region;
Vishal Vermad9b83c72017-08-30 19:36:03 -06001606 nsio = to_nd_namespace_io(&nd_btt->ndns->dev);
1607 btt->phys_bb = &nsio->bb;
Vishal Verma5212e112015-06-25 04:20:32 -04001608
1609 ret = discover_arenas(btt);
1610 if (ret) {
1611 dev_err(dev, "init: error in arena_discover: %d\n", ret);
Dan Williamse32bc722016-03-17 18:23:09 -07001612 return NULL;
Vishal Verma5212e112015-06-25 04:20:32 -04001613 }
1614
Dan Williams58138822015-06-23 20:08:34 -04001615 if (btt->init_state != INIT_READY && nd_region->ro) {
Vishal Vermae6be2dc2017-06-30 18:32:51 -06001616 dev_warn(dev, "%s is read-only, unable to init btt metadata\n",
Dan Williams58138822015-06-23 20:08:34 -04001617 dev_name(&nd_region->dev));
Dan Williamse32bc722016-03-17 18:23:09 -07001618 return NULL;
Dan Williams58138822015-06-23 20:08:34 -04001619 } else if (btt->init_state != INIT_READY) {
Vishal Verma5212e112015-06-25 04:20:32 -04001620 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1621 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1622 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1623 btt->num_arenas, rawsize);
1624
1625 ret = create_arenas(btt);
1626 if (ret) {
1627 dev_info(dev, "init: create_arenas: %d\n", ret);
Dan Williamse32bc722016-03-17 18:23:09 -07001628 return NULL;
Vishal Verma5212e112015-06-25 04:20:32 -04001629 }
1630
1631 ret = btt_meta_init(btt);
1632 if (ret) {
1633 dev_err(dev, "init: error in meta_init: %d\n", ret);
Dan Williamse32bc722016-03-17 18:23:09 -07001634 return NULL;
Vishal Verma5212e112015-06-25 04:20:32 -04001635 }
1636 }
1637
1638 ret = btt_blk_init(btt);
1639 if (ret) {
1640 dev_err(dev, "init: error in blk_init: %d\n", ret);
Dan Williamse32bc722016-03-17 18:23:09 -07001641 return NULL;
Vishal Verma5212e112015-06-25 04:20:32 -04001642 }
1643
1644 btt_debugfs_init(btt);
1645
1646 return btt;
Vishal Verma5212e112015-06-25 04:20:32 -04001647}
1648
1649/**
1650 * btt_fini - de-initialize a BTT
1651 * @btt: the BTT handle that was generated by btt_init
1652 *
1653 * De-initialize a Block Translation Table on device removal
1654 *
1655 * Context:
1656 * Might sleep.
1657 */
1658static void btt_fini(struct btt *btt)
1659{
1660 if (btt) {
1661 btt_blk_cleanup(btt);
1662 free_arenas(btt);
1663 debugfs_remove_recursive(btt->debugfs_dir);
Vishal Verma5212e112015-06-25 04:20:32 -04001664 }
1665}
1666
1667int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1668{
1669 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1670 struct nd_region *nd_region;
Vishal Verma14e49452017-06-28 14:25:00 -06001671 struct btt_sb *btt_sb;
Vishal Verma5212e112015-06-25 04:20:32 -04001672 struct btt *btt;
1673 size_t rawsize;
1674
Dan Williams9dec4892016-04-22 12:26:05 -07001675 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1676 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
Vishal Verma5212e112015-06-25 04:20:32 -04001677 return -ENODEV;
Dan Williams9dec4892016-04-22 12:26:05 -07001678 }
Vishal Verma5212e112015-06-25 04:20:32 -04001679
Vishal Verma14e49452017-06-28 14:25:00 -06001680 btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL);
Christophe Jailleted36b4d2017-08-27 08:30:34 +02001681 if (!btt_sb)
1682 return -ENOMEM;
Vishal Verma14e49452017-06-28 14:25:00 -06001683
1684 /*
1685 * If this returns < 0, that is ok as it just means there wasn't
1686 * an existing BTT, and we're creating a new one. We still need to
1687 * call this as we need the version dependent fields in nd_btt to be
1688 * set correctly based on the holder class
1689 */
1690 nd_btt_version(nd_btt, ndns, btt_sb);
1691
1692 rawsize = nvdimm_namespace_capacity(ndns) - nd_btt->initial_offset;
Vishal Verma5212e112015-06-25 04:20:32 -04001693 if (rawsize < ARENA_MIN_SIZE) {
Dan Williams9dec4892016-04-22 12:26:05 -07001694 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
Vishal Verma14e49452017-06-28 14:25:00 -06001695 dev_name(&ndns->dev),
1696 ARENA_MIN_SIZE + nd_btt->initial_offset);
Vishal Verma5212e112015-06-25 04:20:32 -04001697 return -ENXIO;
1698 }
1699 nd_region = to_nd_region(nd_btt->dev.parent);
1700 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1701 nd_region);
1702 if (!btt)
1703 return -ENOMEM;
1704 nd_btt->btt = btt;
1705
1706 return 0;
1707}
1708EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1709
Dan Williams298f2bc2016-03-15 16:41:04 -07001710int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
Vishal Verma5212e112015-06-25 04:20:32 -04001711{
Vishal Verma5212e112015-06-25 04:20:32 -04001712 struct btt *btt = nd_btt->btt;
1713
1714 btt_fini(btt);
1715 nd_btt->btt = NULL;
1716
1717 return 0;
1718}
1719EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1720
1721static int __init nd_btt_init(void)
1722{
NeilBrownff8e92d2016-03-10 08:59:28 +11001723 int rc = 0;
Vishal Verma5212e112015-06-25 04:20:32 -04001724
1725 debugfs_root = debugfs_create_dir("btt", NULL);
NeilBrownff8e92d2016-03-10 08:59:28 +11001726 if (IS_ERR_OR_NULL(debugfs_root))
Vishal Verma5212e112015-06-25 04:20:32 -04001727 rc = -ENXIO;
Vishal Verma5212e112015-06-25 04:20:32 -04001728
1729 return rc;
1730}
1731
1732static void __exit nd_btt_exit(void)
1733{
1734 debugfs_remove_recursive(debugfs_root);
Vishal Verma5212e112015-06-25 04:20:32 -04001735}
1736
1737MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1738MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>");
1739MODULE_LICENSE("GPL v2");
1740module_init(nd_btt_init);
1741module_exit(nd_btt_exit);