1 #ifndef BLK_THROTTLE_H 1 #ifndef BLK_THROTTLE_H
2 #define BLK_THROTTLE_H 2 #define BLK_THROTTLE_H
3 3
4 #include "blk-cgroup-rwstat.h" 4 #include "blk-cgroup-rwstat.h"
5 5
6 /* 6 /*
7 * To implement hierarchical throttling, throt 7 * To implement hierarchical throttling, throtl_grps form a tree and bios
8 * are dispatched upwards level by level until 8 * are dispatched upwards level by level until they reach the top and get
9 * issued. When dispatching bios from the chi 9 * issued. When dispatching bios from the children and local group at each
10 * level, if the bios are dispatched into a si 10 * level, if the bios are dispatched into a single bio_list, there's a risk
11 * of a local or child group which can queue m 11 * of a local or child group which can queue many bios at once filling up
12 * the list starving others. 12 * the list starving others.
13 * 13 *
14 * To avoid such starvation, dispatched bios a 14 * To avoid such starvation, dispatched bios are queued separately
15 * according to where they came from. When th 15 * according to where they came from. When they are again dispatched to
16 * the parent, they're popped in round-robin o 16 * the parent, they're popped in round-robin order so that no single source
17 * hogs the dispatch window. 17 * hogs the dispatch window.
18 * 18 *
19 * throtl_qnode is used to keep the queued bio 19 * throtl_qnode is used to keep the queued bios separated by their sources.
20 * Bios are queued to throtl_qnode which in tu 20 * Bios are queued to throtl_qnode which in turn is queued to
21 * throtl_service_queue and then dispatched in 21 * throtl_service_queue and then dispatched in round-robin order.
22 * 22 *
23 * It's also used to track the reference count 23 * It's also used to track the reference counts on blkg's. A qnode always
24 * belongs to a throtl_grp and gets queued on 24 * belongs to a throtl_grp and gets queued on itself or the parent, so
25 * incrementing the reference of the associate 25 * incrementing the reference of the associated throtl_grp when a qnode is
26 * queued and decrementing when dequeued is en 26 * queued and decrementing when dequeued is enough to keep the whole blkg
27 * tree pinned while bios are in flight. 27 * tree pinned while bios are in flight.
28 */ 28 */
29 struct throtl_qnode { 29 struct throtl_qnode {
30 struct list_head node; 30 struct list_head node; /* service_queue->queued[] */
31 struct bio_list bios; 31 struct bio_list bios; /* queued bios */
32 struct throtl_grp *tg; 32 struct throtl_grp *tg; /* tg this qnode belongs to */
33 }; 33 };
34 34
35 struct throtl_service_queue { 35 struct throtl_service_queue {
36 struct throtl_service_queue *parent_sq 36 struct throtl_service_queue *parent_sq; /* the parent service_queue */
37 37
38 /* 38 /*
39 * Bios queued directly to this servic 39 * Bios queued directly to this service_queue or dispatched from
40 * children throtl_grp's. 40 * children throtl_grp's.
41 */ 41 */
42 struct list_head queued[2]; 42 struct list_head queued[2]; /* throtl_qnode [READ/WRITE] */
43 unsigned int nr_queued[2]; 43 unsigned int nr_queued[2]; /* number of queued bios */
44 44
45 /* 45 /*
46 * RB tree of active children throtl_g 46 * RB tree of active children throtl_grp's, which are sorted by
47 * their ->disptime. 47 * their ->disptime.
48 */ 48 */
49 struct rb_root_cached pending_tree; 49 struct rb_root_cached pending_tree; /* RB tree of active tgs */
50 unsigned int nr_pending; 50 unsigned int nr_pending; /* # queued in the tree */
51 unsigned long first_pending_ 51 unsigned long first_pending_disptime; /* disptime of the first tg */
52 struct timer_list pending_timer; 52 struct timer_list pending_timer; /* fires on first_pending_disptime */
53 }; 53 };
54 54
55 enum tg_state_flags { 55 enum tg_state_flags {
56 THROTL_TG_PENDING = 1 << 0, 56 THROTL_TG_PENDING = 1 << 0, /* on parent's pending tree */
57 THROTL_TG_WAS_EMPTY = 1 << 1, 57 THROTL_TG_WAS_EMPTY = 1 << 1, /* bio_lists[] became non-empty */
58 THROTL_TG_CANCELING = 1 << 2, 58 THROTL_TG_CANCELING = 1 << 2, /* starts to cancel bio */
59 }; 59 };
60 60
>> 61 enum {
>> 62 LIMIT_LOW,
>> 63 LIMIT_MAX,
>> 64 LIMIT_CNT,
>> 65 };
>> 66
61 struct throtl_grp { 67 struct throtl_grp {
62 /* must be the first member */ 68 /* must be the first member */
63 struct blkg_policy_data pd; 69 struct blkg_policy_data pd;
64 70
65 /* active throtl group service_queue m 71 /* active throtl group service_queue member */
66 struct rb_node rb_node; 72 struct rb_node rb_node;
67 73
68 /* throtl_data this group belongs to * 74 /* throtl_data this group belongs to */
69 struct throtl_data *td; 75 struct throtl_data *td;
70 76
71 /* this group's service queue */ 77 /* this group's service queue */
72 struct throtl_service_queue service_qu 78 struct throtl_service_queue service_queue;
73 79
74 /* 80 /*
75 * qnode_on_self is used when bios are 81 * qnode_on_self is used when bios are directly queued to this
76 * throtl_grp so that local bios compe 82 * throtl_grp so that local bios compete fairly with bios
77 * dispatched from children. qnode_on 83 * dispatched from children. qnode_on_parent is used when bios are
78 * dispatched from this throtl_grp int 84 * dispatched from this throtl_grp into its parent and will compete
79 * with the sibling qnode_on_parents a 85 * with the sibling qnode_on_parents and the parent's
80 * qnode_on_self. 86 * qnode_on_self.
81 */ 87 */
82 struct throtl_qnode qnode_on_self[2]; 88 struct throtl_qnode qnode_on_self[2];
83 struct throtl_qnode qnode_on_parent[2] 89 struct throtl_qnode qnode_on_parent[2];
84 90
85 /* 91 /*
86 * Dispatch time in jiffies. This is t 92 * Dispatch time in jiffies. This is the estimated time when group
87 * will unthrottle and is ready to dis 93 * will unthrottle and is ready to dispatch more bio. It is used as
88 * key to sort active groups in servic 94 * key to sort active groups in service tree.
89 */ 95 */
90 unsigned long disptime; 96 unsigned long disptime;
91 97
92 unsigned int flags; 98 unsigned int flags;
93 99
94 /* are there any throtl rules between 100 /* are there any throtl rules between this group and td? */
95 bool has_rules_bps[2]; 101 bool has_rules_bps[2];
96 bool has_rules_iops[2]; 102 bool has_rules_iops[2];
97 103
98 /* bytes per second rate limits */ !! 104 /* internally used bytes per second rate limits */
99 uint64_t bps[2]; !! 105 uint64_t bps[2][LIMIT_CNT];
100 !! 106 /* user configured bps limits */
101 /* IOPS limits */ !! 107 uint64_t bps_conf[2][LIMIT_CNT];
102 unsigned int iops[2]; !! 108
>> 109 /* internally used IOPS limits */
>> 110 unsigned int iops[2][LIMIT_CNT];
>> 111 /* user configured IOPS limits */
>> 112 unsigned int iops_conf[2][LIMIT_CNT];
103 113
104 /* Number of bytes dispatched in curre 114 /* Number of bytes dispatched in current slice */
105 uint64_t bytes_disp[2]; 115 uint64_t bytes_disp[2];
106 /* Number of bio's dispatched in curre 116 /* Number of bio's dispatched in current slice */
107 unsigned int io_disp[2]; 117 unsigned int io_disp[2];
108 118
109 unsigned long last_low_overflow_time[2 119 unsigned long last_low_overflow_time[2];
110 120
111 uint64_t last_bytes_disp[2]; 121 uint64_t last_bytes_disp[2];
112 unsigned int last_io_disp[2]; 122 unsigned int last_io_disp[2];
113 123
114 /* 124 /*
115 * The following two fields are update 125 * The following two fields are updated when new configuration is
116 * submitted while some bios are still 126 * submitted while some bios are still throttled, they record how many
117 * bytes/ios are waited already in pre 127 * bytes/ios are waited already in previous configuration, and they will
118 * be used to calculate wait time unde 128 * be used to calculate wait time under new configuration.
119 */ 129 */
120 long long carryover_bytes[2]; 130 long long carryover_bytes[2];
121 int carryover_ios[2]; 131 int carryover_ios[2];
122 132
123 unsigned long last_check_time; 133 unsigned long last_check_time;
124 134
>> 135 unsigned long latency_target; /* us */
>> 136 unsigned long latency_target_conf; /* us */
125 /* When did we start a new slice */ 137 /* When did we start a new slice */
126 unsigned long slice_start[2]; 138 unsigned long slice_start[2];
127 unsigned long slice_end[2]; 139 unsigned long slice_end[2];
128 140
>> 141 unsigned long last_finish_time; /* ns / 1024 */
>> 142 unsigned long checked_last_finish_time; /* ns / 1024 */
>> 143 unsigned long avg_idletime; /* ns / 1024 */
>> 144 unsigned long idletime_threshold; /* us */
>> 145 unsigned long idletime_threshold_conf; /* us */
>> 146
>> 147 unsigned int bio_cnt; /* total bios */
>> 148 unsigned int bad_bio_cnt; /* bios exceeding latency threshold */
>> 149 unsigned long bio_cnt_reset_time;
>> 150
129 struct blkg_rwstat stat_bytes; 151 struct blkg_rwstat stat_bytes;
130 struct blkg_rwstat stat_ios; 152 struct blkg_rwstat stat_ios;
131 }; 153 };
132 154
133 extern struct blkcg_policy blkcg_policy_throtl 155 extern struct blkcg_policy blkcg_policy_throtl;
134 156
135 static inline struct throtl_grp *pd_to_tg(stru 157 static inline struct throtl_grp *pd_to_tg(struct blkg_policy_data *pd)
136 { 158 {
137 return pd ? container_of(pd, struct th 159 return pd ? container_of(pd, struct throtl_grp, pd) : NULL;
138 } 160 }
139 161
140 static inline struct throtl_grp *blkg_to_tg(st 162 static inline struct throtl_grp *blkg_to_tg(struct blkcg_gq *blkg)
141 { 163 {
142 return pd_to_tg(blkg_to_pd(blkg, &blkc 164 return pd_to_tg(blkg_to_pd(blkg, &blkcg_policy_throtl));
143 } 165 }
144 166
145 /* 167 /*
146 * Internal throttling interface 168 * Internal throttling interface
147 */ 169 */
148 #ifndef CONFIG_BLK_DEV_THROTTLING 170 #ifndef CONFIG_BLK_DEV_THROTTLING
>> 171 static inline int blk_throtl_init(struct gendisk *disk) { return 0; }
149 static inline void blk_throtl_exit(struct gend 172 static inline void blk_throtl_exit(struct gendisk *disk) { }
>> 173 static inline void blk_throtl_register(struct gendisk *disk) { }
150 static inline bool blk_throtl_bio(struct bio * 174 static inline bool blk_throtl_bio(struct bio *bio) { return false; }
151 static inline void blk_throtl_cancel_bios(stru 175 static inline void blk_throtl_cancel_bios(struct gendisk *disk) { }
152 #else /* CONFIG_BLK_DEV_THROTTLING */ 176 #else /* CONFIG_BLK_DEV_THROTTLING */
>> 177 int blk_throtl_init(struct gendisk *disk);
153 void blk_throtl_exit(struct gendisk *disk); 178 void blk_throtl_exit(struct gendisk *disk);
>> 179 void blk_throtl_register(struct gendisk *disk);
154 bool __blk_throtl_bio(struct bio *bio); 180 bool __blk_throtl_bio(struct bio *bio);
155 void blk_throtl_cancel_bios(struct gendisk *di 181 void blk_throtl_cancel_bios(struct gendisk *disk);
156 182
157 static inline bool blk_throtl_activated(struct <<
158 { <<
159 return q->td != NULL; <<
160 } <<
161 <<
162 static inline bool blk_should_throtl(struct bi 183 static inline bool blk_should_throtl(struct bio *bio)
163 { 184 {
164 struct throtl_grp *tg; !! 185 struct throtl_grp *tg = blkg_to_tg(bio->bi_blkg);
165 int rw = bio_data_dir(bio); 186 int rw = bio_data_dir(bio);
166 187
167 /* <<
168 * This is called under bio_queue_ente <<
169 * the activation of blk-throtl, which <<
170 * blk_mq_freeze_queue(). <<
171 */ <<
172 if (!blk_throtl_activated(bio->bi_bdev <<
173 return false; <<
174 <<
175 tg = blkg_to_tg(bio->bi_blkg); <<
176 if (!cgroup_subsys_on_dfl(io_cgrp_subs 188 if (!cgroup_subsys_on_dfl(io_cgrp_subsys)) {
177 if (!bio_flagged(bio, BIO_CGRO 189 if (!bio_flagged(bio, BIO_CGROUP_ACCT)) {
178 bio_set_flag(bio, BIO_ 190 bio_set_flag(bio, BIO_CGROUP_ACCT);
179 blkg_rwstat_add(&tg->s 191 blkg_rwstat_add(&tg->stat_bytes, bio->bi_opf,
180 bio->b 192 bio->bi_iter.bi_size);
181 } 193 }
182 blkg_rwstat_add(&tg->stat_ios, 194 blkg_rwstat_add(&tg->stat_ios, bio->bi_opf, 1);
183 } 195 }
184 196
185 /* iops limit is always counted */ 197 /* iops limit is always counted */
186 if (tg->has_rules_iops[rw]) 198 if (tg->has_rules_iops[rw])
187 return true; 199 return true;
188 200
189 if (tg->has_rules_bps[rw] && !bio_flag 201 if (tg->has_rules_bps[rw] && !bio_flagged(bio, BIO_BPS_THROTTLED))
190 return true; 202 return true;
191 203
192 return false; 204 return false;
193 } 205 }
194 206
195 static inline bool blk_throtl_bio(struct bio * 207 static inline bool blk_throtl_bio(struct bio *bio)
196 { 208 {
197 209
198 if (!blk_should_throtl(bio)) 210 if (!blk_should_throtl(bio))
199 return false; 211 return false;
200 212
201 return __blk_throtl_bio(bio); 213 return __blk_throtl_bio(bio);
202 } 214 }
203 #endif /* CONFIG_BLK_DEV_THROTTLING */ 215 #endif /* CONFIG_BLK_DEV_THROTTLING */
204 216
205 #endif 217 #endif
206 218
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.