1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * f2fs iostat support
4  *
5  * Copyright 2021 Google LLC
6  * Author: Daeho Jeong <daehojeong@google.com>
7  */
8 
9 #include <linux/fs.h>
10 #include <linux/f2fs_fs.h>
11 #include <linux/seq_file.h>
12 
13 #include "f2fs.h"
14 #include "iostat.h"
15 #include <trace/events/f2fs.h>
16 
17 #define NUM_PREALLOC_IOSTAT_CTXS	128
18 static struct kmem_cache *bio_iostat_ctx_cache;
19 static mempool_t *bio_iostat_ctx_pool;
20 
iostat_info_seq_show(struct seq_file * seq,void * offset)21 int __maybe_unused iostat_info_seq_show(struct seq_file *seq, void *offset)
22 {
23 	struct super_block *sb = seq->private;
24 	struct f2fs_sb_info *sbi = F2FS_SB(sb);
25 	time64_t now = ktime_get_real_seconds();
26 
27 	if (!sbi->iostat_enable)
28 		return 0;
29 
30 	seq_printf(seq, "time:		%-16llu\n", now);
31 
32 	/* print app write IOs */
33 	seq_puts(seq, "[WRITE]\n");
34 	seq_printf(seq, "app buffered:	%-16llu\n",
35 				sbi->rw_iostat[APP_BUFFERED_IO]);
36 	seq_printf(seq, "app direct:	%-16llu\n",
37 				sbi->rw_iostat[APP_DIRECT_IO]);
38 	seq_printf(seq, "app mapped:	%-16llu\n",
39 				sbi->rw_iostat[APP_MAPPED_IO]);
40 
41 	/* print fs write IOs */
42 	seq_printf(seq, "fs data:	%-16llu\n",
43 				sbi->rw_iostat[FS_DATA_IO]);
44 	seq_printf(seq, "fs node:	%-16llu\n",
45 				sbi->rw_iostat[FS_NODE_IO]);
46 	seq_printf(seq, "fs meta:	%-16llu\n",
47 				sbi->rw_iostat[FS_META_IO]);
48 	seq_printf(seq, "fs gc data:	%-16llu\n",
49 				sbi->rw_iostat[FS_GC_DATA_IO]);
50 	seq_printf(seq, "fs gc node:	%-16llu\n",
51 				sbi->rw_iostat[FS_GC_NODE_IO]);
52 	seq_printf(seq, "fs cp data:	%-16llu\n",
53 				sbi->rw_iostat[FS_CP_DATA_IO]);
54 	seq_printf(seq, "fs cp node:	%-16llu\n",
55 				sbi->rw_iostat[FS_CP_NODE_IO]);
56 	seq_printf(seq, "fs cp meta:	%-16llu\n",
57 				sbi->rw_iostat[FS_CP_META_IO]);
58 
59 	/* print app read IOs */
60 	seq_puts(seq, "[READ]\n");
61 	seq_printf(seq, "app buffered:	%-16llu\n",
62 				sbi->rw_iostat[APP_BUFFERED_READ_IO]);
63 	seq_printf(seq, "app direct:	%-16llu\n",
64 				sbi->rw_iostat[APP_DIRECT_READ_IO]);
65 	seq_printf(seq, "app mapped:	%-16llu\n",
66 				sbi->rw_iostat[APP_MAPPED_READ_IO]);
67 
68 	/* print fs read IOs */
69 	seq_printf(seq, "fs data:	%-16llu\n",
70 				sbi->rw_iostat[FS_DATA_READ_IO]);
71 	seq_printf(seq, "fs gc data:	%-16llu\n",
72 				sbi->rw_iostat[FS_GDATA_READ_IO]);
73 	seq_printf(seq, "fs compr_data:	%-16llu\n",
74 				sbi->rw_iostat[FS_CDATA_READ_IO]);
75 	seq_printf(seq, "fs node:	%-16llu\n",
76 				sbi->rw_iostat[FS_NODE_READ_IO]);
77 	seq_printf(seq, "fs meta:	%-16llu\n",
78 				sbi->rw_iostat[FS_META_READ_IO]);
79 
80 	/* print other IOs */
81 	seq_puts(seq, "[OTHER]\n");
82 	seq_printf(seq, "fs discard:	%-16llu\n",
83 				sbi->rw_iostat[FS_DISCARD]);
84 
85 	return 0;
86 }
87 
__record_iostat_latency(struct f2fs_sb_info * sbi)88 static inline void __record_iostat_latency(struct f2fs_sb_info *sbi)
89 {
90 	int io, idx = 0;
91 	unsigned int cnt;
92 	struct f2fs_iostat_latency iostat_lat[MAX_IO_TYPE][NR_PAGE_TYPE];
93 	struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
94 
95 	spin_lock_irq(&sbi->iostat_lat_lock);
96 	for (idx = 0; idx < MAX_IO_TYPE; idx++) {
97 		for (io = 0; io < NR_PAGE_TYPE; io++) {
98 			cnt = io_lat->bio_cnt[idx][io];
99 			iostat_lat[idx][io].peak_lat =
100 			   jiffies_to_msecs(io_lat->peak_lat[idx][io]);
101 			iostat_lat[idx][io].cnt = cnt;
102 			iostat_lat[idx][io].avg_lat = cnt ?
103 			   jiffies_to_msecs(io_lat->sum_lat[idx][io]) / cnt : 0;
104 			io_lat->sum_lat[idx][io] = 0;
105 			io_lat->peak_lat[idx][io] = 0;
106 			io_lat->bio_cnt[idx][io] = 0;
107 		}
108 	}
109 	spin_unlock_irq(&sbi->iostat_lat_lock);
110 
111 	trace_f2fs_iostat_latency(sbi, iostat_lat);
112 }
113 
f2fs_record_iostat(struct f2fs_sb_info * sbi)114 static inline void f2fs_record_iostat(struct f2fs_sb_info *sbi)
115 {
116 	unsigned long long iostat_diff[NR_IO_TYPE];
117 	int i;
118 
119 	if (time_is_after_jiffies(sbi->iostat_next_period))
120 		return;
121 
122 	/* Need double check under the lock */
123 	spin_lock(&sbi->iostat_lock);
124 	if (time_is_after_jiffies(sbi->iostat_next_period)) {
125 		spin_unlock(&sbi->iostat_lock);
126 		return;
127 	}
128 	sbi->iostat_next_period = jiffies +
129 				msecs_to_jiffies(sbi->iostat_period_ms);
130 
131 	for (i = 0; i < NR_IO_TYPE; i++) {
132 		iostat_diff[i] = sbi->rw_iostat[i] -
133 				sbi->prev_rw_iostat[i];
134 		sbi->prev_rw_iostat[i] = sbi->rw_iostat[i];
135 	}
136 	spin_unlock(&sbi->iostat_lock);
137 
138 	trace_f2fs_iostat(sbi, iostat_diff);
139 
140 	__record_iostat_latency(sbi);
141 }
142 
f2fs_reset_iostat(struct f2fs_sb_info * sbi)143 void f2fs_reset_iostat(struct f2fs_sb_info *sbi)
144 {
145 	struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
146 	int i;
147 
148 	spin_lock(&sbi->iostat_lock);
149 	for (i = 0; i < NR_IO_TYPE; i++) {
150 		sbi->rw_iostat[i] = 0;
151 		sbi->prev_rw_iostat[i] = 0;
152 	}
153 	spin_unlock(&sbi->iostat_lock);
154 
155 	spin_lock_irq(&sbi->iostat_lat_lock);
156 	memset(io_lat, 0, sizeof(struct iostat_lat_info));
157 	spin_unlock_irq(&sbi->iostat_lat_lock);
158 }
159 
f2fs_update_iostat(struct f2fs_sb_info * sbi,enum iostat_type type,unsigned long long io_bytes)160 void f2fs_update_iostat(struct f2fs_sb_info *sbi,
161 			enum iostat_type type, unsigned long long io_bytes)
162 {
163 	if (!sbi->iostat_enable)
164 		return;
165 
166 	spin_lock(&sbi->iostat_lock);
167 	sbi->rw_iostat[type] += io_bytes;
168 
169 	if (type == APP_WRITE_IO || type == APP_DIRECT_IO)
170 		sbi->rw_iostat[APP_BUFFERED_IO] =
171 			sbi->rw_iostat[APP_WRITE_IO] -
172 			sbi->rw_iostat[APP_DIRECT_IO];
173 
174 	if (type == APP_READ_IO || type == APP_DIRECT_READ_IO)
175 		sbi->rw_iostat[APP_BUFFERED_READ_IO] =
176 			sbi->rw_iostat[APP_READ_IO] -
177 			sbi->rw_iostat[APP_DIRECT_READ_IO];
178 	spin_unlock(&sbi->iostat_lock);
179 
180 	f2fs_record_iostat(sbi);
181 }
182 
__update_iostat_latency(struct bio_iostat_ctx * iostat_ctx,int rw,bool is_sync)183 static inline void __update_iostat_latency(struct bio_iostat_ctx *iostat_ctx,
184 				int rw, bool is_sync)
185 {
186 	unsigned long ts_diff;
187 	unsigned int iotype = iostat_ctx->type;
188 	unsigned long flags;
189 	struct f2fs_sb_info *sbi = iostat_ctx->sbi;
190 	struct iostat_lat_info *io_lat = sbi->iostat_io_lat;
191 	int idx;
192 
193 	if (!sbi->iostat_enable)
194 		return;
195 
196 	ts_diff = jiffies - iostat_ctx->submit_ts;
197 	if (iotype >= META_FLUSH)
198 		iotype = META;
199 
200 	if (rw == 0) {
201 		idx = READ_IO;
202 	} else {
203 		if (is_sync)
204 			idx = WRITE_SYNC_IO;
205 		else
206 			idx = WRITE_ASYNC_IO;
207 	}
208 
209 	spin_lock_irqsave(&sbi->iostat_lat_lock, flags);
210 	io_lat->sum_lat[idx][iotype] += ts_diff;
211 	io_lat->bio_cnt[idx][iotype]++;
212 	if (ts_diff > io_lat->peak_lat[idx][iotype])
213 		io_lat->peak_lat[idx][iotype] = ts_diff;
214 	spin_unlock_irqrestore(&sbi->iostat_lat_lock, flags);
215 }
216 
iostat_update_and_unbind_ctx(struct bio * bio,int rw)217 void iostat_update_and_unbind_ctx(struct bio *bio, int rw)
218 {
219 	struct bio_iostat_ctx *iostat_ctx = bio->bi_private;
220 	bool is_sync = bio->bi_opf & REQ_SYNC;
221 
222 	if (rw == 0)
223 		bio->bi_private = iostat_ctx->post_read_ctx;
224 	else
225 		bio->bi_private = iostat_ctx->sbi;
226 	__update_iostat_latency(iostat_ctx, rw, is_sync);
227 	mempool_free(iostat_ctx, bio_iostat_ctx_pool);
228 }
229 
iostat_alloc_and_bind_ctx(struct f2fs_sb_info * sbi,struct bio * bio,struct bio_post_read_ctx * ctx)230 void iostat_alloc_and_bind_ctx(struct f2fs_sb_info *sbi,
231 		struct bio *bio, struct bio_post_read_ctx *ctx)
232 {
233 	struct bio_iostat_ctx *iostat_ctx;
234 	/* Due to the mempool, this never fails. */
235 	iostat_ctx = mempool_alloc(bio_iostat_ctx_pool, GFP_NOFS);
236 	iostat_ctx->sbi = sbi;
237 	iostat_ctx->submit_ts = 0;
238 	iostat_ctx->type = 0;
239 	iostat_ctx->post_read_ctx = ctx;
240 	bio->bi_private = iostat_ctx;
241 }
242 
f2fs_init_iostat_processing(void)243 int __init f2fs_init_iostat_processing(void)
244 {
245 	bio_iostat_ctx_cache =
246 		kmem_cache_create("f2fs_bio_iostat_ctx",
247 				  sizeof(struct bio_iostat_ctx), 0, 0, NULL);
248 	if (!bio_iostat_ctx_cache)
249 		goto fail;
250 	bio_iostat_ctx_pool =
251 		mempool_create_slab_pool(NUM_PREALLOC_IOSTAT_CTXS,
252 					 bio_iostat_ctx_cache);
253 	if (!bio_iostat_ctx_pool)
254 		goto fail_free_cache;
255 	return 0;
256 
257 fail_free_cache:
258 	kmem_cache_destroy(bio_iostat_ctx_cache);
259 fail:
260 	return -ENOMEM;
261 }
262 
f2fs_destroy_iostat_processing(void)263 void f2fs_destroy_iostat_processing(void)
264 {
265 	mempool_destroy(bio_iostat_ctx_pool);
266 	kmem_cache_destroy(bio_iostat_ctx_cache);
267 }
268 
f2fs_init_iostat(struct f2fs_sb_info * sbi)269 int f2fs_init_iostat(struct f2fs_sb_info *sbi)
270 {
271 	/* init iostat info */
272 	spin_lock_init(&sbi->iostat_lock);
273 	spin_lock_init(&sbi->iostat_lat_lock);
274 	sbi->iostat_enable = false;
275 	sbi->iostat_period_ms = DEFAULT_IOSTAT_PERIOD_MS;
276 	sbi->iostat_io_lat = f2fs_kzalloc(sbi, sizeof(struct iostat_lat_info),
277 					GFP_KERNEL);
278 	if (!sbi->iostat_io_lat)
279 		return -ENOMEM;
280 
281 	return 0;
282 }
283 
f2fs_destroy_iostat(struct f2fs_sb_info * sbi)284 void f2fs_destroy_iostat(struct f2fs_sb_info *sbi)
285 {
286 	kfree(sbi->iostat_io_lat);
287 }
288