Commit | Line | Data |
---|---|---|
a7b36ee6 JA |
1 | #ifndef BLK_THROTTLE_H |
2 | #define BLK_THROTTLE_H | |
3 | ||
4 | #include "blk-cgroup-rwstat.h" | |
5 | ||
6 | /* | |
7 | * To implement hierarchical throttling, throtl_grps form a tree and bios | |
8 | * are dispatched upwards level by level until they reach the top and get | |
9 | * issued. When dispatching bios from the children and local group at each | |
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 many bios at once filling up | |
12 | * the list starving others. | |
13 | * | |
14 | * To avoid such starvation, dispatched bios are queued separately | |
15 | * according to where they came from. When they are again dispatched to | |
16 | * the parent, they're popped in round-robin order so that no single source | |
17 | * hogs the dispatch window. | |
18 | * | |
19 | * throtl_qnode is used to keep the queued bios separated by their sources. | |
20 | * Bios are queued to throtl_qnode which in turn is queued to | |
21 | * throtl_service_queue and then dispatched in round-robin order. | |
22 | * | |
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 itself or the parent, so | |
25 | * incrementing the reference of the associated throtl_grp when a qnode is | |
26 | * queued and decrementing when dequeued is enough to keep the whole blkg | |
27 | * tree pinned while bios are in flight. | |
28 | */ | |
29 | struct throtl_qnode { | |
30 | struct list_head node; /* service_queue->queued[] */ | |
31 | struct bio_list bios; /* queued bios */ | |
32 | struct throtl_grp *tg; /* tg this qnode belongs to */ | |
33 | }; | |
34 | ||
35 | struct throtl_service_queue { | |
36 | struct throtl_service_queue *parent_sq; /* the parent service_queue */ | |
37 | ||
38 | /* | |
39 | * Bios queued directly to this service_queue or dispatched from | |
40 | * children throtl_grp's. | |
41 | */ | |
42 | struct list_head queued[2]; /* throtl_qnode [READ/WRITE] */ | |
43 | unsigned int nr_queued[2]; /* number of queued bios */ | |
44 | ||
45 | /* | |
46 | * RB tree of active children throtl_grp's, which are sorted by | |
47 | * their ->disptime. | |
48 | */ | |
49 | struct rb_root_cached pending_tree; /* RB tree of active tgs */ | |
50 | unsigned int nr_pending; /* # queued in the tree */ | |
51 | unsigned long first_pending_disptime; /* disptime of the first tg */ | |
52 | struct timer_list pending_timer; /* fires on first_pending_disptime */ | |
53 | }; | |
54 | ||
5a93b602 ML |
55 | enum tg_state_flags { |
56 | THROTL_TG_PENDING = 1 << 0, /* on parent's pending tree */ | |
57 | THROTL_TG_WAS_EMPTY = 1 << 1, /* bio_lists[] became non-empty */ | |
58 | THROTL_TG_HAS_IOPS_LIMIT = 1 << 2, /* tg has iops limit */ | |
8f9e7b65 | 59 | THROTL_TG_CANCELING = 1 << 3, /* starts to cancel bio */ |
5a93b602 ML |
60 | }; |
61 | ||
a7b36ee6 JA |
62 | enum { |
63 | LIMIT_LOW, | |
64 | LIMIT_MAX, | |
65 | LIMIT_CNT, | |
66 | }; | |
67 | ||
68 | struct throtl_grp { | |
69 | /* must be the first member */ | |
70 | struct blkg_policy_data pd; | |
71 | ||
72 | /* active throtl group service_queue member */ | |
73 | struct rb_node rb_node; | |
74 | ||
75 | /* throtl_data this group belongs to */ | |
76 | struct throtl_data *td; | |
77 | ||
78 | /* this group's service queue */ | |
79 | struct throtl_service_queue service_queue; | |
80 | ||
81 | /* | |
82 | * qnode_on_self is used when bios are directly queued to this | |
83 | * throtl_grp so that local bios compete fairly with bios | |
84 | * dispatched from children. qnode_on_parent is used when bios are | |
85 | * dispatched from this throtl_grp into its parent and will compete | |
86 | * with the sibling qnode_on_parents and the parent's | |
87 | * qnode_on_self. | |
88 | */ | |
89 | struct throtl_qnode qnode_on_self[2]; | |
90 | struct throtl_qnode qnode_on_parent[2]; | |
91 | ||
92 | /* | |
93 | * Dispatch time in jiffies. This is the estimated time when group | |
94 | * will unthrottle and is ready to dispatch more bio. It is used as | |
95 | * key to sort active groups in service tree. | |
96 | */ | |
97 | unsigned long disptime; | |
98 | ||
99 | unsigned int flags; | |
100 | ||
101 | /* are there any throtl rules between this group and td? */ | |
102 | bool has_rules[2]; | |
103 | ||
104 | /* internally used bytes per second rate limits */ | |
105 | uint64_t bps[2][LIMIT_CNT]; | |
106 | /* user configured bps limits */ | |
107 | uint64_t bps_conf[2][LIMIT_CNT]; | |
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]; | |
113 | ||
114 | /* Number of bytes dispatched in current slice */ | |
115 | uint64_t bytes_disp[2]; | |
116 | /* Number of bio's dispatched in current slice */ | |
117 | unsigned int io_disp[2]; | |
118 | ||
119 | unsigned long last_low_overflow_time[2]; | |
120 | ||
121 | uint64_t last_bytes_disp[2]; | |
122 | unsigned int last_io_disp[2]; | |
123 | ||
124 | unsigned long last_check_time; | |
125 | ||
126 | unsigned long latency_target; /* us */ | |
127 | unsigned long latency_target_conf; /* us */ | |
128 | /* When did we start a new slice */ | |
129 | unsigned long slice_start[2]; | |
130 | unsigned long slice_end[2]; | |
131 | ||
132 | unsigned long last_finish_time; /* ns / 1024 */ | |
133 | unsigned long checked_last_finish_time; /* ns / 1024 */ | |
134 | unsigned long avg_idletime; /* ns / 1024 */ | |
135 | unsigned long idletime_threshold; /* us */ | |
136 | unsigned long idletime_threshold_conf; /* us */ | |
137 | ||
138 | unsigned int bio_cnt; /* total bios */ | |
139 | unsigned int bad_bio_cnt; /* bios exceeding latency threshold */ | |
140 | unsigned long bio_cnt_reset_time; | |
141 | ||
a7b36ee6 JA |
142 | struct blkg_rwstat stat_bytes; |
143 | struct blkg_rwstat stat_ios; | |
144 | }; | |
145 | ||
146 | extern struct blkcg_policy blkcg_policy_throtl; | |
147 | ||
148 | static inline struct throtl_grp *pd_to_tg(struct blkg_policy_data *pd) | |
149 | { | |
150 | return pd ? container_of(pd, struct throtl_grp, pd) : NULL; | |
151 | } | |
152 | ||
153 | static inline struct throtl_grp *blkg_to_tg(struct blkcg_gq *blkg) | |
154 | { | |
155 | return pd_to_tg(blkg_to_pd(blkg, &blkcg_policy_throtl)); | |
156 | } | |
157 | ||
158 | /* | |
159 | * Internal throttling interface | |
160 | */ | |
161 | #ifndef CONFIG_BLK_DEV_THROTTLING | |
162 | static inline int blk_throtl_init(struct request_queue *q) { return 0; } | |
163 | static inline void blk_throtl_exit(struct request_queue *q) { } | |
164 | static inline void blk_throtl_register_queue(struct request_queue *q) { } | |
a7b36ee6 | 165 | static inline bool blk_throtl_bio(struct bio *bio) { return false; } |
8f9e7b65 | 166 | static inline void blk_throtl_cancel_bios(struct request_queue *q) { } |
a7b36ee6 JA |
167 | #else /* CONFIG_BLK_DEV_THROTTLING */ |
168 | int blk_throtl_init(struct request_queue *q); | |
169 | void blk_throtl_exit(struct request_queue *q); | |
170 | void blk_throtl_register_queue(struct request_queue *q); | |
a7b36ee6 | 171 | bool __blk_throtl_bio(struct bio *bio); |
8f9e7b65 | 172 | void blk_throtl_cancel_bios(struct request_queue *q); |
a7b36ee6 JA |
173 | static inline bool blk_throtl_bio(struct bio *bio) |
174 | { | |
175 | struct throtl_grp *tg = blkg_to_tg(bio->bi_blkg); | |
176 | ||
5a93b602 ML |
177 | /* no need to throttle bps any more if the bio has been throttled */ |
178 | if (bio_flagged(bio, BIO_THROTTLED) && | |
179 | !(tg->flags & THROTL_TG_HAS_IOPS_LIMIT)) | |
180 | return false; | |
181 | ||
a7b36ee6 JA |
182 | if (!tg->has_rules[bio_data_dir(bio)]) |
183 | return false; | |
184 | ||
185 | return __blk_throtl_bio(bio); | |
186 | } | |
187 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
188 | ||
189 | #endif |