2 * Copyright (C) 2016 CNEX Labs
3 * Initial release: Javier Gonzalez <javier@cnexlabs.com>
4 * Matias Bjorling <matias@cnexlabs.com>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License version
8 * 2 as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * pblk-rl.c - pblk's rate limiter for user I/O
21 static void pblk_rl_kick_u_timer(struct pblk_rl
*rl
)
23 mod_timer(&rl
->u_timer
, jiffies
+ msecs_to_jiffies(5000));
26 int pblk_rl_is_limit(struct pblk_rl
*rl
)
30 rb_space
= atomic_read(&rl
->rb_space
);
32 return (rb_space
== 0);
35 int pblk_rl_user_may_insert(struct pblk_rl
*rl
, int nr_entries
)
37 int rb_user_cnt
= atomic_read(&rl
->rb_user_cnt
);
38 int rb_space
= atomic_read(&rl
->rb_space
);
40 if (unlikely(rb_space
>= 0) && (rb_space
- nr_entries
< 0))
43 if (rb_user_cnt
>= rl
->rb_user_max
)
44 return NVM_IO_REQUEUE
;
49 void pblk_rl_inserted(struct pblk_rl
*rl
, int nr_entries
)
51 int rb_space
= atomic_read(&rl
->rb_space
);
53 if (unlikely(rb_space
>= 0))
54 atomic_sub(nr_entries
, &rl
->rb_space
);
57 int pblk_rl_gc_may_insert(struct pblk_rl
*rl
, int nr_entries
)
59 int rb_gc_cnt
= atomic_read(&rl
->rb_gc_cnt
);
62 /* If there is no user I/O let GC take over space on the write buffer */
63 rb_user_active
= READ_ONCE(rl
->rb_user_active
);
64 return (!(rb_gc_cnt
>= rl
->rb_gc_max
&& rb_user_active
));
67 void pblk_rl_user_in(struct pblk_rl
*rl
, int nr_entries
)
69 atomic_add(nr_entries
, &rl
->rb_user_cnt
);
71 /* Release user I/O state. Protect from GC */
72 smp_store_release(&rl
->rb_user_active
, 1);
73 pblk_rl_kick_u_timer(rl
);
76 void pblk_rl_werr_line_in(struct pblk_rl
*rl
)
78 atomic_inc(&rl
->werr_lines
);
81 void pblk_rl_werr_line_out(struct pblk_rl
*rl
)
83 atomic_dec(&rl
->werr_lines
);
86 void pblk_rl_gc_in(struct pblk_rl
*rl
, int nr_entries
)
88 atomic_add(nr_entries
, &rl
->rb_gc_cnt
);
91 void pblk_rl_out(struct pblk_rl
*rl
, int nr_user
, int nr_gc
)
93 atomic_sub(nr_user
, &rl
->rb_user_cnt
);
94 atomic_sub(nr_gc
, &rl
->rb_gc_cnt
);
97 unsigned long pblk_rl_nr_free_blks(struct pblk_rl
*rl
)
99 return atomic_read(&rl
->free_blocks
);
102 unsigned long pblk_rl_nr_user_free_blks(struct pblk_rl
*rl
)
104 return atomic_read(&rl
->free_user_blocks
);
107 static void __pblk_rl_update_rates(struct pblk_rl
*rl
,
108 unsigned long free_blocks
)
110 struct pblk
*pblk
= container_of(rl
, struct pblk
, rl
);
111 int max
= rl
->rb_budget
;
112 int werr_gc_needed
= atomic_read(&rl
->werr_lines
);
114 if (free_blocks
>= rl
->high
) {
115 if (werr_gc_needed
) {
116 /* Allocate a small budget for recovering
117 * lines with write errors
119 rl
->rb_gc_max
= 1 << rl
->rb_windows_pw
;
120 rl
->rb_user_max
= max
- rl
->rb_gc_max
;
121 rl
->rb_state
= PBLK_RL_WERR
;
123 rl
->rb_user_max
= max
;
125 rl
->rb_state
= PBLK_RL_OFF
;
127 } else if (free_blocks
< rl
->high
) {
128 int shift
= rl
->high_pw
- rl
->rb_windows_pw
;
129 int user_windows
= free_blocks
>> shift
;
130 int user_max
= user_windows
<< PBLK_MAX_REQ_ADDRS_PW
;
132 rl
->rb_user_max
= user_max
;
133 rl
->rb_gc_max
= max
- user_max
;
135 if (free_blocks
<= rl
->rsv_blocks
) {
140 /* In the worst case, we will need to GC lines in the low list
141 * (high valid sector count). If there are lines to GC on high
142 * or mid lists, these will be prioritized
144 rl
->rb_state
= PBLK_RL_LOW
;
147 if (rl
->rb_state
!= PBLK_RL_OFF
)
148 pblk_gc_should_start(pblk
);
150 pblk_gc_should_stop(pblk
);
153 void pblk_rl_update_rates(struct pblk_rl
*rl
)
155 __pblk_rl_update_rates(rl
, pblk_rl_nr_user_free_blks(rl
));
158 void pblk_rl_free_lines_inc(struct pblk_rl
*rl
, struct pblk_line
*line
)
160 int blk_in_line
= atomic_read(&line
->blk_in_line
);
163 atomic_add(blk_in_line
, &rl
->free_blocks
);
164 free_blocks
= atomic_add_return(blk_in_line
, &rl
->free_user_blocks
);
166 __pblk_rl_update_rates(rl
, free_blocks
);
169 void pblk_rl_free_lines_dec(struct pblk_rl
*rl
, struct pblk_line
*line
,
172 int blk_in_line
= atomic_read(&line
->blk_in_line
);
175 atomic_sub(blk_in_line
, &rl
->free_blocks
);
178 free_blocks
= atomic_sub_return(blk_in_line
,
179 &rl
->free_user_blocks
);
181 free_blocks
= atomic_read(&rl
->free_user_blocks
);
183 __pblk_rl_update_rates(rl
, free_blocks
);
186 int pblk_rl_high_thrs(struct pblk_rl
*rl
)
191 int pblk_rl_max_io(struct pblk_rl
*rl
)
193 return rl
->rb_max_io
;
196 static void pblk_rl_u_timer(struct timer_list
*t
)
198 struct pblk_rl
*rl
= from_timer(rl
, t
, u_timer
);
200 /* Release user I/O state. Protect from GC */
201 smp_store_release(&rl
->rb_user_active
, 0);
204 void pblk_rl_free(struct pblk_rl
*rl
)
206 del_timer(&rl
->u_timer
);
209 void pblk_rl_init(struct pblk_rl
*rl
, int budget
)
211 struct pblk
*pblk
= container_of(rl
, struct pblk
, rl
);
212 struct nvm_tgt_dev
*dev
= pblk
->dev
;
213 struct nvm_geo
*geo
= &dev
->geo
;
214 struct pblk_line_mgmt
*l_mg
= &pblk
->l_mg
;
215 struct pblk_line_meta
*lm
= &pblk
->lm
;
216 int min_blocks
= lm
->blk_per_line
* PBLK_GC_RSV_LINE
;
217 int sec_meta
, blk_meta
;
219 unsigned int rb_windows
;
221 /* Consider sectors used for metadata */
222 sec_meta
= (lm
->smeta_sec
+ lm
->emeta_sec
[0]) * l_mg
->nr_free_lines
;
223 blk_meta
= DIV_ROUND_UP(sec_meta
, geo
->clba
);
225 rl
->high
= pblk
->op_blks
- blk_meta
- lm
->blk_per_line
;
226 rl
->high_pw
= get_count_order(rl
->high
);
228 rl
->rsv_blocks
= min_blocks
;
230 /* This will always be a power-of-2 */
231 rb_windows
= budget
/ PBLK_MAX_REQ_ADDRS
;
232 rl
->rb_windows_pw
= get_count_order(rb_windows
);
234 /* To start with, all buffer is available to user I/O writers */
235 rl
->rb_budget
= budget
;
236 rl
->rb_user_max
= budget
;
237 rl
->rb_max_io
= budget
>> 1;
239 rl
->rb_state
= PBLK_RL_HIGH
;
241 atomic_set(&rl
->rb_user_cnt
, 0);
242 atomic_set(&rl
->rb_gc_cnt
, 0);
243 atomic_set(&rl
->rb_space
, -1);
244 atomic_set(&rl
->werr_lines
, 0);
246 timer_setup(&rl
->u_timer
, pblk_rl_u_timer
, 0);
248 rl
->rb_user_active
= 0;
249 rl
->rb_gc_active
= 0;