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 | ||
55 | enum { | |
56 | LIMIT_LOW, | |
57 | LIMIT_MAX, | |
58 | LIMIT_CNT, | |
59 | }; | |
60 | ||
61 | struct throtl_grp { | |
62 | /* must be the first member */ | |
63 | struct blkg_policy_data pd; | |
64 | ||
65 | /* active throtl group service_queue member */ | |
66 | struct rb_node rb_node; | |
67 | ||
68 | /* throtl_data this group belongs to */ | |
69 | struct throtl_data *td; | |
70 | ||
71 | /* this group's service queue */ | |
72 | struct throtl_service_queue service_queue; | |
73 | ||
74 | /* | |
75 | * qnode_on_self is used when bios are directly queued to this | |
76 | * throtl_grp so that local bios compete fairly with bios | |
77 | * dispatched from children. qnode_on_parent is used when bios are | |
78 | * dispatched from this throtl_grp into its parent and will compete | |
79 | * with the sibling qnode_on_parents and the parent's | |
80 | * qnode_on_self. | |
81 | */ | |
82 | struct throtl_qnode qnode_on_self[2]; | |
83 | struct throtl_qnode qnode_on_parent[2]; | |
84 | ||
85 | /* | |
86 | * Dispatch time in jiffies. This is the estimated time when group | |
87 | * will unthrottle and is ready to dispatch more bio. It is used as | |
88 | * key to sort active groups in service tree. | |
89 | */ | |
90 | unsigned long disptime; | |
91 | ||
92 | unsigned int flags; | |
93 | ||
94 | /* are there any throtl rules between this group and td? */ | |
95 | bool has_rules[2]; | |
96 | ||
97 | /* internally used bytes per second rate limits */ | |
98 | uint64_t bps[2][LIMIT_CNT]; | |
99 | /* user configured bps limits */ | |
100 | uint64_t bps_conf[2][LIMIT_CNT]; | |
101 | ||
102 | /* internally used IOPS limits */ | |
103 | unsigned int iops[2][LIMIT_CNT]; | |
104 | /* user configured IOPS limits */ | |
105 | unsigned int iops_conf[2][LIMIT_CNT]; | |
106 | ||
107 | /* Number of bytes dispatched in current slice */ | |
108 | uint64_t bytes_disp[2]; | |
109 | /* Number of bio's dispatched in current slice */ | |
110 | unsigned int io_disp[2]; | |
111 | ||
112 | unsigned long last_low_overflow_time[2]; | |
113 | ||
114 | uint64_t last_bytes_disp[2]; | |
115 | unsigned int last_io_disp[2]; | |
116 | ||
117 | unsigned long last_check_time; | |
118 | ||
119 | unsigned long latency_target; /* us */ | |
120 | unsigned long latency_target_conf; /* us */ | |
121 | /* When did we start a new slice */ | |
122 | unsigned long slice_start[2]; | |
123 | unsigned long slice_end[2]; | |
124 | ||
125 | unsigned long last_finish_time; /* ns / 1024 */ | |
126 | unsigned long checked_last_finish_time; /* ns / 1024 */ | |
127 | unsigned long avg_idletime; /* ns / 1024 */ | |
128 | unsigned long idletime_threshold; /* us */ | |
129 | unsigned long idletime_threshold_conf; /* us */ | |
130 | ||
131 | unsigned int bio_cnt; /* total bios */ | |
132 | unsigned int bad_bio_cnt; /* bios exceeding latency threshold */ | |
133 | unsigned long bio_cnt_reset_time; | |
134 | ||
135 | atomic_t io_split_cnt[2]; | |
136 | atomic_t last_io_split_cnt[2]; | |
137 | ||
138 | struct blkg_rwstat stat_bytes; | |
139 | struct blkg_rwstat stat_ios; | |
140 | }; | |
141 | ||
142 | extern struct blkcg_policy blkcg_policy_throtl; | |
143 | ||
144 | static inline struct throtl_grp *pd_to_tg(struct blkg_policy_data *pd) | |
145 | { | |
146 | return pd ? container_of(pd, struct throtl_grp, pd) : NULL; | |
147 | } | |
148 | ||
149 | static inline struct throtl_grp *blkg_to_tg(struct blkcg_gq *blkg) | |
150 | { | |
151 | return pd_to_tg(blkg_to_pd(blkg, &blkcg_policy_throtl)); | |
152 | } | |
153 | ||
154 | /* | |
155 | * Internal throttling interface | |
156 | */ | |
157 | #ifndef CONFIG_BLK_DEV_THROTTLING | |
158 | static inline int blk_throtl_init(struct request_queue *q) { return 0; } | |
159 | static inline void blk_throtl_exit(struct request_queue *q) { } | |
160 | static inline void blk_throtl_register_queue(struct request_queue *q) { } | |
161 | static inline void blk_throtl_charge_bio_split(struct bio *bio) { } | |
162 | static inline bool blk_throtl_bio(struct bio *bio) { return false; } | |
163 | #else /* CONFIG_BLK_DEV_THROTTLING */ | |
164 | int blk_throtl_init(struct request_queue *q); | |
165 | void blk_throtl_exit(struct request_queue *q); | |
166 | void blk_throtl_register_queue(struct request_queue *q); | |
167 | void blk_throtl_charge_bio_split(struct bio *bio); | |
168 | bool __blk_throtl_bio(struct bio *bio); | |
169 | static inline bool blk_throtl_bio(struct bio *bio) | |
170 | { | |
171 | struct throtl_grp *tg = blkg_to_tg(bio->bi_blkg); | |
172 | ||
173 | if (bio_flagged(bio, BIO_THROTTLED)) | |
174 | return false; | |
175 | if (!tg->has_rules[bio_data_dir(bio)]) | |
176 | return false; | |
177 | ||
178 | return __blk_throtl_bio(bio); | |
179 | } | |
180 | #endif /* CONFIG_BLK_DEV_THROTTLING */ | |
181 | ||
182 | #endif |