Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | * CFQ, or complete fairness queueing, disk scheduler. |
| 3 | * |
| 4 | * Based on ideas from a previously unfinished io |
| 5 | * scheduler (round robin per-process disk scheduling) and Andrea Arcangeli. |
| 6 | * |
Jens Axboe | 0fe2347 | 2006-09-04 15:41:16 +0200 | [diff] [blame] | 7 | * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 8 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | #include <linux/module.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 10 | #include <linux/slab.h> |
Al Viro | 1cc9be6 | 2006-03-18 12:29:52 -0500 | [diff] [blame] | 11 | #include <linux/blkdev.h> |
| 12 | #include <linux/elevator.h> |
Randy Dunlap | ad5ebd2 | 2009-11-11 13:47:45 +0100 | [diff] [blame] | 13 | #include <linux/jiffies.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/rbtree.h> |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 15 | #include <linux/ioprio.h> |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 16 | #include <linux/blktrace_api.h> |
Tejun Heo | 6e736be | 2011-12-14 00:33:38 +0100 | [diff] [blame] | 17 | #include "blk.h" |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 18 | #include "blk-cgroup.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 20 | static struct blkio_policy_type blkio_policy_cfq; |
| 21 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | /* |
| 23 | * tunables |
| 24 | */ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 25 | /* max queue in one round of service */ |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 26 | static const int cfq_quantum = 8; |
Arjan van de Ven | 6410009 | 2006-01-06 09:46:02 +0100 | [diff] [blame] | 27 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 28 | /* maximum backwards seek, in KiB */ |
| 29 | static const int cfq_back_max = 16 * 1024; |
| 30 | /* penalty of a backwards seek */ |
| 31 | static const int cfq_back_penalty = 2; |
Arjan van de Ven | 6410009 | 2006-01-06 09:46:02 +0100 | [diff] [blame] | 32 | static const int cfq_slice_sync = HZ / 10; |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 33 | static int cfq_slice_async = HZ / 25; |
Arjan van de Ven | 6410009 | 2006-01-06 09:46:02 +0100 | [diff] [blame] | 34 | static const int cfq_slice_async_rq = 2; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 35 | static int cfq_slice_idle = HZ / 125; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 36 | static int cfq_group_idle = HZ / 125; |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 37 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
| 38 | static const int cfq_hist_divisor = 4; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 39 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 40 | /* |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 41 | * offset from end of service tree |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 42 | */ |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 43 | #define CFQ_IDLE_DELAY (HZ / 5) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 44 | |
| 45 | /* |
| 46 | * below this threshold, we consider thinktime immediate |
| 47 | */ |
| 48 | #define CFQ_MIN_TT (2) |
| 49 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 50 | #define CFQ_SLICE_SCALE (5) |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 51 | #define CFQ_HW_QUEUE_MIN (5) |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 52 | #define CFQ_SERVICE_SHIFT 12 |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 53 | |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 54 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 55 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
Corrado Zoccolo | 41647e7 | 2010-02-27 19:45:40 +0100 | [diff] [blame] | 56 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 57 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 58 | |
Tejun Heo | a612fdd | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 59 | #define RQ_CIC(rq) icq_to_cic((rq)->elv.icq) |
| 60 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elv.priv[0]) |
| 61 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elv.priv[1]) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 63 | static struct kmem_cache *cfq_pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 65 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
| 66 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 67 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
| 68 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 69 | #define sample_valid(samples) ((samples) > 80) |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 70 | #define rb_entry_cfqg(node) rb_entry((node), struct cfq_group, rb_node) |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 71 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 72 | struct cfq_ttime { |
| 73 | unsigned long last_end_request; |
| 74 | |
| 75 | unsigned long ttime_total; |
| 76 | unsigned long ttime_samples; |
| 77 | unsigned long ttime_mean; |
| 78 | }; |
| 79 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 80 | /* |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 81 | * Most of our rbtree usage is for sorting with min extraction, so |
| 82 | * if we cache the leftmost node we don't have to walk down the tree |
| 83 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should |
| 84 | * move this into the elevator for the rq sorting as well. |
| 85 | */ |
| 86 | struct cfq_rb_root { |
| 87 | struct rb_root rb; |
| 88 | struct rb_node *left; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 89 | unsigned count; |
Richard Kennedy | 73e9ffd | 2010-03-01 10:50:20 +0100 | [diff] [blame] | 90 | unsigned total_weight; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 91 | u64 min_vdisktime; |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 92 | struct cfq_ttime ttime; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 93 | }; |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 94 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
| 95 | .ttime = {.last_end_request = jiffies,},} |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 96 | |
| 97 | /* |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 98 | * Per process-grouping structure |
| 99 | */ |
| 100 | struct cfq_queue { |
| 101 | /* reference count */ |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 102 | int ref; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 103 | /* various state flags, see below */ |
| 104 | unsigned int flags; |
| 105 | /* parent cfq_data */ |
| 106 | struct cfq_data *cfqd; |
| 107 | /* service_tree member */ |
| 108 | struct rb_node rb_node; |
| 109 | /* service_tree key */ |
| 110 | unsigned long rb_key; |
| 111 | /* prio tree member */ |
| 112 | struct rb_node p_node; |
| 113 | /* prio tree root we belong to, if any */ |
| 114 | struct rb_root *p_root; |
| 115 | /* sorted list of pending requests */ |
| 116 | struct rb_root sort_list; |
| 117 | /* if fifo isn't expired, next request to serve */ |
| 118 | struct request *next_rq; |
| 119 | /* requests queued in sort_list */ |
| 120 | int queued[2]; |
| 121 | /* currently allocated requests */ |
| 122 | int allocated[2]; |
| 123 | /* fifo list of requests in sort_list */ |
| 124 | struct list_head fifo; |
| 125 | |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 126 | /* time when queue got scheduled in to dispatch first request. */ |
| 127 | unsigned long dispatch_start; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 128 | unsigned int allocated_slice; |
Richard Kennedy | c4081ba | 2010-02-22 13:49:24 +0100 | [diff] [blame] | 129 | unsigned int slice_dispatch; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 130 | /* time when first request from queue completed and slice started. */ |
| 131 | unsigned long slice_start; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 132 | unsigned long slice_end; |
| 133 | long slice_resid; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 134 | |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 135 | /* pending priority requests */ |
| 136 | int prio_pending; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 137 | /* number of requests that are on the dispatch list or inside driver */ |
| 138 | int dispatched; |
| 139 | |
| 140 | /* io prio of this group */ |
| 141 | unsigned short ioprio, org_ioprio; |
Justin TerAvest | 4aede84 | 2011-07-12 08:31:45 +0200 | [diff] [blame] | 142 | unsigned short ioprio_class; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 143 | |
Richard Kennedy | c4081ba | 2010-02-22 13:49:24 +0100 | [diff] [blame] | 144 | pid_t pid; |
| 145 | |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 146 | u32 seek_history; |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 147 | sector_t last_request_pos; |
| 148 | |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 149 | struct cfq_rb_root *service_tree; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 150 | struct cfq_queue *new_cfqq; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 151 | struct cfq_group *cfqg; |
Vivek Goyal | c4e7893 | 2010-08-23 12:25:03 +0200 | [diff] [blame] | 152 | /* Number of sectors dispatched from queue in single dispatch round */ |
| 153 | unsigned long nr_sectors; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 154 | }; |
| 155 | |
| 156 | /* |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 157 | * First index in the service_trees. |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 158 | * IDLE is handled separately, so it has negative index |
| 159 | */ |
| 160 | enum wl_prio_t { |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 161 | BE_WORKLOAD = 0, |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 162 | RT_WORKLOAD = 1, |
| 163 | IDLE_WORKLOAD = 2, |
Vivek Goyal | b462732 | 2010-10-22 09:48:43 +0200 | [diff] [blame] | 164 | CFQ_PRIO_NR, |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 165 | }; |
| 166 | |
| 167 | /* |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 168 | * Second index in the service_trees. |
| 169 | */ |
| 170 | enum wl_type_t { |
| 171 | ASYNC_WORKLOAD = 0, |
| 172 | SYNC_NOIDLE_WORKLOAD = 1, |
| 173 | SYNC_WORKLOAD = 2 |
| 174 | }; |
| 175 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 176 | /* This is per cgroup per device grouping structure */ |
| 177 | struct cfq_group { |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 178 | /* group service_tree member */ |
| 179 | struct rb_node rb_node; |
| 180 | |
| 181 | /* group service_tree key */ |
| 182 | u64 vdisktime; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 183 | unsigned int weight; |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 184 | unsigned int new_weight; |
| 185 | bool needs_update; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 186 | |
| 187 | /* number of cfqq currently on this group */ |
| 188 | int nr_cfqq; |
| 189 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 190 | /* |
Kyungmin Park | 4495a7d | 2011-05-31 10:04:09 +0200 | [diff] [blame] | 191 | * Per group busy queues average. Useful for workload slice calc. We |
Vivek Goyal | b462732 | 2010-10-22 09:48:43 +0200 | [diff] [blame] | 192 | * create the array for each prio class but at run time it is used |
| 193 | * only for RT and BE class and slot for IDLE class remains unused. |
| 194 | * This is primarily done to avoid confusion and a gcc warning. |
| 195 | */ |
| 196 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; |
| 197 | /* |
| 198 | * rr lists of queues with requests. We maintain service trees for |
| 199 | * RT and BE classes. These trees are subdivided in subclasses |
| 200 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE |
| 201 | * class there is no subclassification and all the cfq queues go on |
| 202 | * a single tree service_tree_idle. |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 203 | * Counts are embedded in the cfq_rb_root |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 204 | */ |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 205 | struct cfq_rb_root service_trees[2][3]; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 206 | struct cfq_rb_root service_tree_idle; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 207 | |
| 208 | unsigned long saved_workload_slice; |
| 209 | enum wl_type_t saved_workload; |
| 210 | enum wl_prio_t saved_serving_prio; |
Tejun Heo | 4eef304 | 2012-03-05 13:15:18 -0800 | [diff] [blame] | 211 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 212 | /* number of requests that are on the dispatch list or inside driver */ |
| 213 | int dispatched; |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 214 | struct cfq_ttime ttime; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 215 | }; |
| 216 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 217 | struct cfq_io_cq { |
| 218 | struct io_cq icq; /* must be the first member */ |
| 219 | struct cfq_queue *cfqq[2]; |
| 220 | struct cfq_ttime ttime; |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 221 | int ioprio; /* the current ioprio */ |
| 222 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 223 | uint64_t blkcg_id; /* the current blkcg ID */ |
| 224 | #endif |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 225 | }; |
| 226 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 227 | /* |
| 228 | * Per block device queue structure |
| 229 | */ |
| 230 | struct cfq_data { |
| 231 | struct request_queue *queue; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 232 | /* Root service tree for cfq_groups */ |
| 233 | struct cfq_rb_root grp_service_tree; |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 234 | struct cfq_group *root_group; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 235 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 236 | /* |
| 237 | * The priority currently being served |
| 238 | */ |
| 239 | enum wl_prio_t serving_prio; |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 240 | enum wl_type_t serving_type; |
| 241 | unsigned long workload_expires; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 242 | struct cfq_group *serving_group; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 243 | |
| 244 | /* |
| 245 | * Each priority tree is sorted by next_request position. These |
| 246 | * trees are used when determining if two or more queues are |
| 247 | * interleaving requests (see cfq_close_cooperator). |
| 248 | */ |
| 249 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; |
| 250 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 251 | unsigned int busy_queues; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 252 | unsigned int busy_sync_queues; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 253 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 254 | int rq_in_driver; |
| 255 | int rq_in_flight[2]; |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 256 | |
| 257 | /* |
| 258 | * queue-depth detection |
| 259 | */ |
| 260 | int rq_queued; |
Jens Axboe | 25776e3 | 2006-06-01 10:12:26 +0200 | [diff] [blame] | 261 | int hw_tag; |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 262 | /* |
| 263 | * hw_tag can be |
| 264 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) |
| 265 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) |
| 266 | * 0 => no NCQ |
| 267 | */ |
| 268 | int hw_tag_est_depth; |
| 269 | unsigned int hw_tag_samples; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 270 | |
| 271 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 272 | * idle window management |
| 273 | */ |
| 274 | struct timer_list idle_slice_timer; |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 275 | struct work_struct unplug_work; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 276 | |
| 277 | struct cfq_queue *active_queue; |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 278 | struct cfq_io_cq *active_cic; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 279 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 280 | /* |
| 281 | * async queue for each priority case |
| 282 | */ |
| 283 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; |
| 284 | struct cfq_queue *async_idle_cfqq; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 285 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 286 | sector_t last_position; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | /* |
| 289 | * tunables, see top of file |
| 290 | */ |
| 291 | unsigned int cfq_quantum; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 292 | unsigned int cfq_fifo_expire[2]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | unsigned int cfq_back_penalty; |
| 294 | unsigned int cfq_back_max; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 295 | unsigned int cfq_slice[2]; |
| 296 | unsigned int cfq_slice_async_rq; |
| 297 | unsigned int cfq_slice_idle; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 298 | unsigned int cfq_group_idle; |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 299 | unsigned int cfq_latency; |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 300 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 301 | /* |
| 302 | * Fallback dummy cfqq for extreme OOM conditions |
| 303 | */ |
| 304 | struct cfq_queue oom_cfqq; |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 305 | |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 306 | unsigned long last_delayed_sync; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | }; |
| 308 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 309 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
| 310 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 311 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
| 312 | enum wl_prio_t prio, |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 313 | enum wl_type_t type) |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 314 | { |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 315 | if (!cfqg) |
| 316 | return NULL; |
| 317 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 318 | if (prio == IDLE_WORKLOAD) |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 319 | return &cfqg->service_tree_idle; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 320 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 321 | return &cfqg->service_trees[prio][type]; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 322 | } |
| 323 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 324 | enum cfqq_state_flags { |
Jens Axboe | b0b8d749 | 2007-01-19 11:35:30 +1100 | [diff] [blame] | 325 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
| 326 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 327 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
Jens Axboe | b0b8d749 | 2007-01-19 11:35:30 +1100 | [diff] [blame] | 328 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
Jens Axboe | b0b8d749 | 2007-01-19 11:35:30 +1100 | [diff] [blame] | 329 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
| 330 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ |
| 331 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 332 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 333 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 334 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 335 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 336 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 337 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 338 | }; |
| 339 | |
| 340 | #define CFQ_CFQQ_FNS(name) \ |
| 341 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ |
| 342 | { \ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 343 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 344 | } \ |
| 345 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ |
| 346 | { \ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 347 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 348 | } \ |
| 349 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ |
| 350 | { \ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 351 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 352 | } |
| 353 | |
| 354 | CFQ_CFQQ_FNS(on_rr); |
| 355 | CFQ_CFQQ_FNS(wait_request); |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 356 | CFQ_CFQQ_FNS(must_dispatch); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 357 | CFQ_CFQQ_FNS(must_alloc_slice); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 358 | CFQ_CFQQ_FNS(fifo_expire); |
| 359 | CFQ_CFQQ_FNS(idle_window); |
| 360 | CFQ_CFQQ_FNS(prio_changed); |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 361 | CFQ_CFQQ_FNS(slice_new); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 362 | CFQ_CFQQ_FNS(sync); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 363 | CFQ_CFQQ_FNS(coop); |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 364 | CFQ_CFQQ_FNS(split_coop); |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 365 | CFQ_CFQQ_FNS(deep); |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 366 | CFQ_CFQQ_FNS(wait_busy); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 367 | #undef CFQ_CFQQ_FNS |
| 368 | |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 369 | #if defined(CONFIG_CFQ_GROUP_IOSCHED) && defined(CONFIG_DEBUG_BLK_CGROUP) |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 370 | |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 371 | /* blkg state flags */ |
| 372 | enum blkg_state_flags { |
| 373 | BLKG_waiting = 0, |
| 374 | BLKG_idling, |
| 375 | BLKG_empty, |
| 376 | }; |
| 377 | |
| 378 | #define BLKG_FLAG_FNS(name) \ |
| 379 | static inline void blkio_mark_blkg_##name( \ |
| 380 | struct blkio_group_stats *stats) \ |
| 381 | { \ |
| 382 | stats->flags |= (1 << BLKG_##name); \ |
| 383 | } \ |
| 384 | static inline void blkio_clear_blkg_##name( \ |
| 385 | struct blkio_group_stats *stats) \ |
| 386 | { \ |
| 387 | stats->flags &= ~(1 << BLKG_##name); \ |
| 388 | } \ |
| 389 | static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \ |
| 390 | { \ |
| 391 | return (stats->flags & (1 << BLKG_##name)) != 0; \ |
| 392 | } \ |
| 393 | |
| 394 | BLKG_FLAG_FNS(waiting) |
| 395 | BLKG_FLAG_FNS(idling) |
| 396 | BLKG_FLAG_FNS(empty) |
| 397 | #undef BLKG_FLAG_FNS |
| 398 | |
| 399 | /* This should be called with the queue_lock held. */ |
| 400 | static void blkio_update_group_wait_time(struct blkio_group_stats *stats) |
| 401 | { |
| 402 | unsigned long long now; |
| 403 | |
| 404 | if (!blkio_blkg_waiting(stats)) |
| 405 | return; |
| 406 | |
| 407 | now = sched_clock(); |
| 408 | if (time_after64(now, stats->start_group_wait_time)) |
| 409 | blkg_stat_add(&stats->group_wait_time, |
| 410 | now - stats->start_group_wait_time); |
| 411 | blkio_clear_blkg_waiting(stats); |
| 412 | } |
| 413 | |
| 414 | /* This should be called with the queue_lock held. */ |
| 415 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, |
| 416 | struct blkio_policy_type *pol, |
| 417 | struct blkio_group *curr_blkg) |
| 418 | { |
| 419 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
| 420 | |
| 421 | if (blkio_blkg_waiting(&pd->stats)) |
| 422 | return; |
| 423 | if (blkg == curr_blkg) |
| 424 | return; |
| 425 | pd->stats.start_group_wait_time = sched_clock(); |
| 426 | blkio_mark_blkg_waiting(&pd->stats); |
| 427 | } |
| 428 | |
| 429 | /* This should be called with the queue_lock held. */ |
| 430 | static void blkio_end_empty_time(struct blkio_group_stats *stats) |
| 431 | { |
| 432 | unsigned long long now; |
| 433 | |
| 434 | if (!blkio_blkg_empty(stats)) |
| 435 | return; |
| 436 | |
| 437 | now = sched_clock(); |
| 438 | if (time_after64(now, stats->start_empty_time)) |
| 439 | blkg_stat_add(&stats->empty_time, |
| 440 | now - stats->start_empty_time); |
| 441 | blkio_clear_blkg_empty(stats); |
| 442 | } |
| 443 | |
| 444 | static void cfq_blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
| 445 | struct blkio_policy_type *pol, |
| 446 | unsigned long dequeue) |
| 447 | { |
| 448 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
| 449 | |
| 450 | lockdep_assert_held(blkg->q->queue_lock); |
| 451 | |
| 452 | blkg_stat_add(&pd->stats.dequeue, dequeue); |
| 453 | } |
| 454 | |
| 455 | static void cfq_blkiocg_set_start_empty_time(struct blkio_group *blkg, |
| 456 | struct blkio_policy_type *pol) |
| 457 | { |
| 458 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 459 | |
| 460 | lockdep_assert_held(blkg->q->queue_lock); |
| 461 | |
| 462 | if (blkg_rwstat_sum(&stats->queued)) |
| 463 | return; |
| 464 | |
| 465 | /* |
| 466 | * group is already marked empty. This can happen if cfqq got new |
| 467 | * request in parent group and moved to this group while being added |
| 468 | * to service tree. Just ignore the event and move on. |
| 469 | */ |
| 470 | if (blkio_blkg_empty(stats)) |
| 471 | return; |
| 472 | |
| 473 | stats->start_empty_time = sched_clock(); |
| 474 | blkio_mark_blkg_empty(stats); |
| 475 | } |
| 476 | |
| 477 | static void cfq_blkiocg_update_idle_time_stats(struct blkio_group *blkg, |
| 478 | struct blkio_policy_type *pol) |
| 479 | { |
| 480 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 481 | |
| 482 | lockdep_assert_held(blkg->q->queue_lock); |
| 483 | |
| 484 | if (blkio_blkg_idling(stats)) { |
| 485 | unsigned long long now = sched_clock(); |
| 486 | |
| 487 | if (time_after64(now, stats->start_idle_time)) |
| 488 | blkg_stat_add(&stats->idle_time, |
| 489 | now - stats->start_idle_time); |
| 490 | blkio_clear_blkg_idling(stats); |
| 491 | } |
| 492 | } |
| 493 | |
| 494 | static void cfq_blkiocg_update_set_idle_time_stats(struct blkio_group *blkg, |
| 495 | struct blkio_policy_type *pol) |
| 496 | { |
| 497 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 498 | |
| 499 | lockdep_assert_held(blkg->q->queue_lock); |
| 500 | BUG_ON(blkio_blkg_idling(stats)); |
| 501 | |
| 502 | stats->start_idle_time = sched_clock(); |
| 503 | blkio_mark_blkg_idling(stats); |
| 504 | } |
| 505 | |
| 506 | static void cfq_blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg, |
| 507 | struct blkio_policy_type *pol) |
| 508 | { |
| 509 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 510 | |
| 511 | lockdep_assert_held(blkg->q->queue_lock); |
| 512 | |
| 513 | blkg_stat_add(&stats->avg_queue_size_sum, |
| 514 | blkg_rwstat_sum(&stats->queued)); |
| 515 | blkg_stat_add(&stats->avg_queue_size_samples, 1); |
| 516 | blkio_update_group_wait_time(stats); |
| 517 | } |
| 518 | |
| 519 | #else /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ |
| 520 | |
| 521 | static void blkio_set_start_group_wait_time(struct blkio_group *blkg, |
| 522 | struct blkio_policy_type *pol, |
| 523 | struct blkio_group *curr_blkg) { } |
| 524 | static void blkio_end_empty_time(struct blkio_group_stats *stats) { } |
| 525 | static void cfq_blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
| 526 | struct blkio_policy_type *pol, |
| 527 | unsigned long dequeue) { } |
| 528 | static void cfq_blkiocg_set_start_empty_time(struct blkio_group *blkg, |
| 529 | struct blkio_policy_type *pol) { } |
| 530 | static void cfq_blkiocg_update_idle_time_stats(struct blkio_group *blkg, |
| 531 | struct blkio_policy_type *pol) { } |
| 532 | static void cfq_blkiocg_update_set_idle_time_stats(struct blkio_group *blkg, |
| 533 | struct blkio_policy_type *pol) { } |
| 534 | static void cfq_blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg, |
| 535 | struct blkio_policy_type *pol) { } |
| 536 | |
| 537 | #endif /* CONFIG_CFQ_GROUP_IOSCHED && CONFIG_DEBUG_BLK_CGROUP */ |
| 538 | |
| 539 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 540 | |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 541 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) |
| 542 | { |
| 543 | return blkg_to_pdata(blkg, &blkio_policy_cfq); |
| 544 | } |
| 545 | |
| 546 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) |
| 547 | { |
Tejun Heo | aaec55a | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 548 | return pdata_to_blkg(cfqg); |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 549 | } |
| 550 | |
| 551 | static inline void cfqg_get(struct cfq_group *cfqg) |
| 552 | { |
| 553 | return blkg_get(cfqg_to_blkg(cfqg)); |
| 554 | } |
| 555 | |
| 556 | static inline void cfqg_put(struct cfq_group *cfqg) |
| 557 | { |
| 558 | return blkg_put(cfqg_to_blkg(cfqg)); |
| 559 | } |
| 560 | |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 561 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
| 562 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ |
| 563 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 564 | blkg_path(cfqg_to_blkg((cfqq)->cfqg)), ##args) |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 565 | |
| 566 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ |
| 567 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 568 | blkg_path(cfqg_to_blkg((cfqg))), ##args) \ |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 569 | |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 570 | static inline void cfq_blkiocg_update_io_add_stats(struct blkio_group *blkg, |
| 571 | struct blkio_policy_type *pol, |
| 572 | struct blkio_group *curr_blkg, |
| 573 | bool direction, bool sync) |
| 574 | { |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 575 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 576 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 577 | |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 578 | lockdep_assert_held(blkg->q->queue_lock); |
| 579 | |
| 580 | blkg_rwstat_add(&stats->queued, rw, 1); |
| 581 | blkio_end_empty_time(stats); |
| 582 | blkio_set_start_group_wait_time(blkg, pol, curr_blkg); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 583 | } |
| 584 | |
| 585 | static inline void cfq_blkiocg_update_timeslice_used(struct blkio_group *blkg, |
| 586 | struct blkio_policy_type *pol, unsigned long time, |
| 587 | unsigned long unaccounted_time) |
| 588 | { |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 589 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 590 | |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 591 | lockdep_assert_held(blkg->q->queue_lock); |
| 592 | |
| 593 | blkg_stat_add(&stats->time, time); |
| 594 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 595 | blkg_stat_add(&stats->unaccounted_time, unaccounted_time); |
| 596 | #endif |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 597 | } |
| 598 | |
| 599 | static inline void cfq_blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
| 600 | struct blkio_policy_type *pol, bool direction, |
| 601 | bool sync) |
| 602 | { |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 603 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 604 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
| 605 | |
| 606 | lockdep_assert_held(blkg->q->queue_lock); |
| 607 | |
| 608 | blkg_rwstat_add(&stats->queued, rw, -1); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 609 | } |
| 610 | |
| 611 | static inline void cfq_blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
| 612 | struct blkio_policy_type *pol, bool direction, |
| 613 | bool sync) |
| 614 | { |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 615 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 616 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 617 | |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 618 | lockdep_assert_held(blkg->q->queue_lock); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 619 | |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 620 | blkg_rwstat_add(&stats->merged, rw, 1); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 621 | } |
| 622 | |
| 623 | static inline void cfq_blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
| 624 | struct blkio_policy_type *pol, uint64_t bytes, |
| 625 | bool direction, bool sync) |
| 626 | { |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 627 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
| 628 | struct blkg_policy_data *pd = blkg->pd[pol->plid]; |
| 629 | struct blkio_group_stats_cpu *stats_cpu; |
| 630 | unsigned long flags; |
| 631 | |
| 632 | /* If per cpu stats are not allocated yet, don't do any accounting. */ |
| 633 | if (pd->stats_cpu == NULL) |
| 634 | return; |
| 635 | |
| 636 | /* |
| 637 | * Disabling interrupts to provide mutual exclusion between two |
| 638 | * writes on same cpu. It probably is not needed for 64bit. Not |
| 639 | * optimizing that case yet. |
| 640 | */ |
| 641 | local_irq_save(flags); |
| 642 | |
| 643 | stats_cpu = this_cpu_ptr(pd->stats_cpu); |
| 644 | |
| 645 | blkg_stat_add(&stats_cpu->sectors, bytes >> 9); |
| 646 | blkg_rwstat_add(&stats_cpu->serviced, rw, 1); |
| 647 | blkg_rwstat_add(&stats_cpu->service_bytes, rw, bytes); |
| 648 | |
| 649 | local_irq_restore(flags); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 650 | } |
| 651 | |
| 652 | static inline void cfq_blkiocg_update_completion_stats(struct blkio_group *blkg, |
| 653 | struct blkio_policy_type *pol, uint64_t start_time, |
| 654 | uint64_t io_start_time, bool direction, bool sync) |
| 655 | { |
Tejun Heo | 629ed0b | 2012-04-01 14:38:44 -0700 | [diff] [blame^] | 656 | struct blkio_group_stats *stats = &blkg->pd[pol->plid]->stats; |
| 657 | unsigned long long now = sched_clock(); |
| 658 | int rw = (direction ? REQ_WRITE : 0) | (sync ? REQ_SYNC : 0); |
| 659 | |
| 660 | lockdep_assert_held(blkg->q->queue_lock); |
| 661 | |
| 662 | if (time_after64(now, io_start_time)) |
| 663 | blkg_rwstat_add(&stats->service_time, rw, now - io_start_time); |
| 664 | if (time_after64(io_start_time, start_time)) |
| 665 | blkg_rwstat_add(&stats->wait_time, rw, |
| 666 | io_start_time - start_time); |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 667 | } |
| 668 | |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 669 | #else /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 670 | |
| 671 | static inline struct cfq_group *blkg_to_cfqg(struct blkio_group *blkg) { return NULL; } |
| 672 | static inline struct blkio_group *cfqg_to_blkg(struct cfq_group *cfqg) { return NULL; } |
| 673 | static inline void cfqg_get(struct cfq_group *cfqg) { } |
| 674 | static inline void cfqg_put(struct cfq_group *cfqg) { } |
| 675 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 676 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
| 677 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) |
Kyungmin Park | 4495a7d | 2011-05-31 10:04:09 +0200 | [diff] [blame] | 678 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 679 | |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 680 | static inline void cfq_blkiocg_update_io_add_stats(struct blkio_group *blkg, |
| 681 | struct blkio_policy_type *pol, |
| 682 | struct blkio_group *curr_blkg, bool direction, |
| 683 | bool sync) { } |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 684 | static inline void cfq_blkiocg_update_timeslice_used(struct blkio_group *blkg, |
| 685 | struct blkio_policy_type *pol, unsigned long time, |
| 686 | unsigned long unaccounted_time) { } |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 687 | static inline void cfq_blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
| 688 | struct blkio_policy_type *pol, bool direction, |
| 689 | bool sync) { } |
| 690 | static inline void cfq_blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
| 691 | struct blkio_policy_type *pol, bool direction, |
| 692 | bool sync) { } |
Tejun Heo | 2ce4d50 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 693 | static inline void cfq_blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
| 694 | struct blkio_policy_type *pol, uint64_t bytes, |
| 695 | bool direction, bool sync) { } |
| 696 | static inline void cfq_blkiocg_update_completion_stats(struct blkio_group *blkg, |
| 697 | struct blkio_policy_type *pol, uint64_t start_time, |
| 698 | uint64_t io_start_time, bool direction, bool sync) { } |
| 699 | |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 700 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 701 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 702 | #define cfq_log(cfqd, fmt, args...) \ |
| 703 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) |
| 704 | |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 705 | /* Traverses through cfq group service trees */ |
| 706 | #define for_each_cfqg_st(cfqg, i, j, st) \ |
| 707 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ |
| 708 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ |
| 709 | : &cfqg->service_tree_idle; \ |
| 710 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ |
| 711 | (i == IDLE_WORKLOAD && j == 0); \ |
| 712 | j++, st = i < IDLE_WORKLOAD ? \ |
| 713 | &cfqg->service_trees[i][j]: NULL) \ |
| 714 | |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 715 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
| 716 | struct cfq_ttime *ttime, bool group_idle) |
| 717 | { |
| 718 | unsigned long slice; |
| 719 | if (!sample_valid(ttime->ttime_samples)) |
| 720 | return false; |
| 721 | if (group_idle) |
| 722 | slice = cfqd->cfq_group_idle; |
| 723 | else |
| 724 | slice = cfqd->cfq_slice_idle; |
| 725 | return ttime->ttime_mean > slice; |
| 726 | } |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 727 | |
Vivek Goyal | 02b3508 | 2010-08-23 12:23:53 +0200 | [diff] [blame] | 728 | static inline bool iops_mode(struct cfq_data *cfqd) |
| 729 | { |
| 730 | /* |
| 731 | * If we are not idling on queues and it is a NCQ drive, parallel |
| 732 | * execution of requests is on and measuring time is not possible |
| 733 | * in most of the cases until and unless we drive shallower queue |
| 734 | * depths and that becomes a performance bottleneck. In such cases |
| 735 | * switch to start providing fairness in terms of number of IOs. |
| 736 | */ |
| 737 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) |
| 738 | return true; |
| 739 | else |
| 740 | return false; |
| 741 | } |
| 742 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 743 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
| 744 | { |
| 745 | if (cfq_class_idle(cfqq)) |
| 746 | return IDLE_WORKLOAD; |
| 747 | if (cfq_class_rt(cfqq)) |
| 748 | return RT_WORKLOAD; |
| 749 | return BE_WORKLOAD; |
| 750 | } |
| 751 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 752 | |
| 753 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) |
| 754 | { |
| 755 | if (!cfq_cfqq_sync(cfqq)) |
| 756 | return ASYNC_WORKLOAD; |
| 757 | if (!cfq_cfqq_idle_window(cfqq)) |
| 758 | return SYNC_NOIDLE_WORKLOAD; |
| 759 | return SYNC_WORKLOAD; |
| 760 | } |
| 761 | |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 762 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
| 763 | struct cfq_data *cfqd, |
| 764 | struct cfq_group *cfqg) |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 765 | { |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 766 | if (wl == IDLE_WORKLOAD) |
| 767 | return cfqg->service_tree_idle.count; |
| 768 | |
| 769 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
| 770 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count |
| 771 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 772 | } |
| 773 | |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 774 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
| 775 | struct cfq_group *cfqg) |
| 776 | { |
| 777 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count |
| 778 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; |
| 779 | } |
| 780 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 781 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
Tejun Heo | 4f85cb9 | 2012-03-05 13:15:28 -0800 | [diff] [blame] | 782 | static struct cfq_queue *cfq_get_queue(struct cfq_data *cfqd, bool is_sync, |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 783 | struct cfq_io_cq *cic, struct bio *bio, |
Tejun Heo | 4f85cb9 | 2012-03-05 13:15:28 -0800 | [diff] [blame] | 784 | gfp_t gfp_mask); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 785 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 786 | static inline struct cfq_io_cq *icq_to_cic(struct io_cq *icq) |
| 787 | { |
| 788 | /* cic->icq is the first member, %NULL will convert to %NULL */ |
| 789 | return container_of(icq, struct cfq_io_cq, icq); |
| 790 | } |
| 791 | |
Tejun Heo | 47fdd4c | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 792 | static inline struct cfq_io_cq *cfq_cic_lookup(struct cfq_data *cfqd, |
| 793 | struct io_context *ioc) |
| 794 | { |
| 795 | if (ioc) |
| 796 | return icq_to_cic(ioc_lookup_icq(ioc, cfqd->queue)); |
| 797 | return NULL; |
| 798 | } |
| 799 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 800 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_cq *cic, bool is_sync) |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 801 | { |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 802 | return cic->cfqq[is_sync]; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 803 | } |
| 804 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 805 | static inline void cic_set_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq, |
| 806 | bool is_sync) |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 807 | { |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 808 | cic->cfqq[is_sync] = cfqq; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 809 | } |
| 810 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 811 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_cq *cic) |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 812 | { |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 813 | return cic->icq.q->elevator->elevator_data; |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 814 | } |
| 815 | |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 816 | /* |
| 817 | * We regard a request as SYNC, if it's either a read or has the SYNC bit |
| 818 | * set (in which case it could also be direct WRITE). |
| 819 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 820 | static inline bool cfq_bio_sync(struct bio *bio) |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 821 | { |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 822 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 823 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 824 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | /* |
Andrew Morton | 99f95e5 | 2005-06-27 20:14:05 -0700 | [diff] [blame] | 826 | * scheduler run of queue, if there are requests pending and no one in the |
| 827 | * driver that will restart queueing |
| 828 | */ |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 829 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
Andrew Morton | 99f95e5 | 2005-06-27 20:14:05 -0700 | [diff] [blame] | 830 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 831 | if (cfqd->busy_queues) { |
| 832 | cfq_log(cfqd, "schedule dispatch"); |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 833 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 834 | } |
Andrew Morton | 99f95e5 | 2005-06-27 20:14:05 -0700 | [diff] [blame] | 835 | } |
| 836 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | /* |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 838 | * Scale schedule slice based on io priority. Use the sync time slice only |
| 839 | * if a queue is marked sync and has sync io queued. A sync queue with async |
| 840 | * io only, should not get full sync slice length. |
| 841 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 842 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 843 | unsigned short prio) |
| 844 | { |
| 845 | const int base_slice = cfqd->cfq_slice[sync]; |
| 846 | |
| 847 | WARN_ON(prio >= IOPRIO_BE_NR); |
| 848 | |
| 849 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); |
| 850 | } |
| 851 | |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 852 | static inline int |
| 853 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 854 | { |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 855 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 856 | } |
| 857 | |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 858 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
| 859 | { |
| 860 | u64 d = delta << CFQ_SERVICE_SHIFT; |
| 861 | |
| 862 | d = d * BLKIO_WEIGHT_DEFAULT; |
| 863 | do_div(d, cfqg->weight); |
| 864 | return d; |
| 865 | } |
| 866 | |
| 867 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) |
| 868 | { |
| 869 | s64 delta = (s64)(vdisktime - min_vdisktime); |
| 870 | if (delta > 0) |
| 871 | min_vdisktime = vdisktime; |
| 872 | |
| 873 | return min_vdisktime; |
| 874 | } |
| 875 | |
| 876 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) |
| 877 | { |
| 878 | s64 delta = (s64)(vdisktime - min_vdisktime); |
| 879 | if (delta < 0) |
| 880 | min_vdisktime = vdisktime; |
| 881 | |
| 882 | return min_vdisktime; |
| 883 | } |
| 884 | |
| 885 | static void update_min_vdisktime(struct cfq_rb_root *st) |
| 886 | { |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 887 | struct cfq_group *cfqg; |
| 888 | |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 889 | if (st->left) { |
| 890 | cfqg = rb_entry_cfqg(st->left); |
Gui Jianfeng | a603271 | 2011-03-07 09:28:09 +0100 | [diff] [blame] | 891 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
| 892 | cfqg->vdisktime); |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 893 | } |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 894 | } |
| 895 | |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 896 | /* |
| 897 | * get averaged number of queues of RT/BE priority. |
| 898 | * average is updated, with a formula that gives more weight to higher numbers, |
| 899 | * to quickly follows sudden increases and decrease slowly |
| 900 | */ |
| 901 | |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 902 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
| 903 | struct cfq_group *cfqg, bool rt) |
Jens Axboe | 5869619c | 2009-10-28 09:27:07 +0100 | [diff] [blame] | 904 | { |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 905 | unsigned min_q, max_q; |
| 906 | unsigned mult = cfq_hist_divisor - 1; |
| 907 | unsigned round = cfq_hist_divisor / 2; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 908 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 909 | |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 910 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
| 911 | max_q = max(cfqg->busy_queues_avg[rt], busy); |
| 912 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 913 | cfq_hist_divisor; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 914 | return cfqg->busy_queues_avg[rt]; |
| 915 | } |
| 916 | |
| 917 | static inline unsigned |
| 918 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) |
| 919 | { |
| 920 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 921 | |
| 922 | return cfq_target_latency * cfqg->weight / st->total_weight; |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 923 | } |
| 924 | |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 925 | static inline unsigned |
Vivek Goyal | ba5bd52 | 2011-01-19 08:25:02 -0700 | [diff] [blame] | 926 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 927 | { |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 928 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
| 929 | if (cfqd->cfq_latency) { |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 930 | /* |
| 931 | * interested queues (we consider only the ones with the same |
| 932 | * priority class in the cfq group) |
| 933 | */ |
| 934 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, |
| 935 | cfq_class_rt(cfqq)); |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 936 | unsigned sync_slice = cfqd->cfq_slice[1]; |
| 937 | unsigned expect_latency = sync_slice * iq; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 938 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
| 939 | |
| 940 | if (expect_latency > group_slice) { |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 941 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
| 942 | /* scale low_slice according to IO priority |
| 943 | * and sync vs async */ |
| 944 | unsigned low_slice = |
| 945 | min(slice, base_low_slice * slice / sync_slice); |
| 946 | /* the adapted slice value is scaled to fit all iqs |
| 947 | * into the target latency */ |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 948 | slice = max(slice * group_slice / expect_latency, |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 949 | low_slice); |
| 950 | } |
| 951 | } |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 952 | return slice; |
| 953 | } |
| 954 | |
| 955 | static inline void |
| 956 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 957 | { |
Vivek Goyal | ba5bd52 | 2011-01-19 08:25:02 -0700 | [diff] [blame] | 958 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 959 | |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 960 | cfqq->slice_start = jiffies; |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 961 | cfqq->slice_end = jiffies + slice; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 962 | cfqq->allocated_slice = slice; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 963 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 964 | } |
| 965 | |
| 966 | /* |
| 967 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end |
| 968 | * isn't valid until the first request from the dispatch is activated |
| 969 | * and the slice time set. |
| 970 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 971 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 972 | { |
| 973 | if (cfq_cfqq_slice_new(cfqq)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 974 | return false; |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 975 | if (time_before(jiffies, cfqq->slice_end)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 976 | return false; |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 977 | |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 978 | return true; |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 979 | } |
| 980 | |
| 981 | /* |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 982 | * Lifted from AS - choose which of rq1 and rq2 that is best served now. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 983 | * We choose the request that is closest to the head right now. Distance |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 984 | * behind the head is penalized and only allowed to a certain extent. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 985 | */ |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 986 | static struct request * |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 987 | cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2, sector_t last) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 988 | { |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 989 | sector_t s1, s2, d1 = 0, d2 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 990 | unsigned long back_max; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 991 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
| 992 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ |
| 993 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 994 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 995 | if (rq1 == NULL || rq1 == rq2) |
| 996 | return rq2; |
| 997 | if (rq2 == NULL) |
| 998 | return rq1; |
Jens Axboe | 9c2c38a | 2005-08-24 14:57:54 +0200 | [diff] [blame] | 999 | |
Namhyung Kim | 229836b | 2011-05-24 10:23:21 +0200 | [diff] [blame] | 1000 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
| 1001 | return rq_is_sync(rq1) ? rq1 : rq2; |
| 1002 | |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 1003 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
| 1004 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 1005 | |
Tejun Heo | 83096eb | 2009-05-07 22:24:39 +0900 | [diff] [blame] | 1006 | s1 = blk_rq_pos(rq1); |
| 1007 | s2 = blk_rq_pos(rq2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1008 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1009 | /* |
| 1010 | * by definition, 1KiB is 2 sectors |
| 1011 | */ |
| 1012 | back_max = cfqd->cfq_back_max * 2; |
| 1013 | |
| 1014 | /* |
| 1015 | * Strict one way elevator _except_ in the case where we allow |
| 1016 | * short backward seeks which are biased as twice the cost of a |
| 1017 | * similar forward seek. |
| 1018 | */ |
| 1019 | if (s1 >= last) |
| 1020 | d1 = s1 - last; |
| 1021 | else if (s1 + back_max >= last) |
| 1022 | d1 = (last - s1) * cfqd->cfq_back_penalty; |
| 1023 | else |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1024 | wrap |= CFQ_RQ1_WRAP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1025 | |
| 1026 | if (s2 >= last) |
| 1027 | d2 = s2 - last; |
| 1028 | else if (s2 + back_max >= last) |
| 1029 | d2 = (last - s2) * cfqd->cfq_back_penalty; |
| 1030 | else |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1031 | wrap |= CFQ_RQ2_WRAP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | |
| 1033 | /* Found required data */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1034 | |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1035 | /* |
| 1036 | * By doing switch() on the bit mask "wrap" we avoid having to |
| 1037 | * check two variables for all permutations: --> faster! |
| 1038 | */ |
| 1039 | switch (wrap) { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1040 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1041 | if (d1 < d2) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1042 | return rq1; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1043 | else if (d2 < d1) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1044 | return rq2; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1045 | else { |
| 1046 | if (s1 >= s2) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1047 | return rq1; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1048 | else |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1049 | return rq2; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1050 | } |
| 1051 | |
| 1052 | case CFQ_RQ2_WRAP: |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1053 | return rq1; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1054 | case CFQ_RQ1_WRAP: |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1055 | return rq2; |
| 1056 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 1057 | default: |
| 1058 | /* |
| 1059 | * Since both rqs are wrapped, |
| 1060 | * start with the one that's further behind head |
| 1061 | * (--> only *one* back seek required), |
| 1062 | * since back seek takes more time than forward. |
| 1063 | */ |
| 1064 | if (s1 <= s2) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1065 | return rq1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1066 | else |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1067 | return rq2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1068 | } |
| 1069 | } |
| 1070 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1071 | /* |
| 1072 | * The below is leftmost cache rbtree addon |
| 1073 | */ |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1074 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1075 | { |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 1076 | /* Service tree is empty */ |
| 1077 | if (!root->count) |
| 1078 | return NULL; |
| 1079 | |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1080 | if (!root->left) |
| 1081 | root->left = rb_first(&root->rb); |
| 1082 | |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1083 | if (root->left) |
| 1084 | return rb_entry(root->left, struct cfq_queue, rb_node); |
| 1085 | |
| 1086 | return NULL; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1087 | } |
| 1088 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1089 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
| 1090 | { |
| 1091 | if (!root->left) |
| 1092 | root->left = rb_first(&root->rb); |
| 1093 | |
| 1094 | if (root->left) |
| 1095 | return rb_entry_cfqg(root->left); |
| 1096 | |
| 1097 | return NULL; |
| 1098 | } |
| 1099 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1100 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
| 1101 | { |
| 1102 | rb_erase(n, root); |
| 1103 | RB_CLEAR_NODE(n); |
| 1104 | } |
| 1105 | |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1106 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
| 1107 | { |
| 1108 | if (root->left == n) |
| 1109 | root->left = NULL; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1110 | rb_erase_init(n, &root->rb); |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1111 | --root->count; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1112 | } |
| 1113 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1114 | /* |
| 1115 | * would be nice to take fifo expire time into account as well |
| 1116 | */ |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1117 | static struct request * |
| 1118 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 1119 | struct request *last) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | { |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1121 | struct rb_node *rbnext = rb_next(&last->rb_node); |
| 1122 | struct rb_node *rbprev = rb_prev(&last->rb_node); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1123 | struct request *next = NULL, *prev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1124 | |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1125 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1126 | |
| 1127 | if (rbprev) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1128 | prev = rb_entry_rq(rbprev); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1129 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1130 | if (rbnext) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1131 | next = rb_entry_rq(rbnext); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1132 | else { |
| 1133 | rbnext = rb_first(&cfqq->sort_list); |
| 1134 | if (rbnext && rbnext != &last->rb_node) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1135 | next = rb_entry_rq(rbnext); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1136 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1137 | |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 1138 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | } |
| 1140 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1141 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
| 1142 | struct cfq_queue *cfqq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | { |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1144 | /* |
| 1145 | * just an approximation, should be ok. |
| 1146 | */ |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 1147 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
Jens Axboe | 464191c | 2009-11-30 09:38:13 +0100 | [diff] [blame] | 1148 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1149 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1150 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1151 | static inline s64 |
| 1152 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1153 | { |
| 1154 | return cfqg->vdisktime - st->min_vdisktime; |
| 1155 | } |
| 1156 | |
| 1157 | static void |
| 1158 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1159 | { |
| 1160 | struct rb_node **node = &st->rb.rb_node; |
| 1161 | struct rb_node *parent = NULL; |
| 1162 | struct cfq_group *__cfqg; |
| 1163 | s64 key = cfqg_key(st, cfqg); |
| 1164 | int left = 1; |
| 1165 | |
| 1166 | while (*node != NULL) { |
| 1167 | parent = *node; |
| 1168 | __cfqg = rb_entry_cfqg(parent); |
| 1169 | |
| 1170 | if (key < cfqg_key(st, __cfqg)) |
| 1171 | node = &parent->rb_left; |
| 1172 | else { |
| 1173 | node = &parent->rb_right; |
| 1174 | left = 0; |
| 1175 | } |
| 1176 | } |
| 1177 | |
| 1178 | if (left) |
| 1179 | st->left = &cfqg->rb_node; |
| 1180 | |
| 1181 | rb_link_node(&cfqg->rb_node, parent, node); |
| 1182 | rb_insert_color(&cfqg->rb_node, &st->rb); |
| 1183 | } |
| 1184 | |
| 1185 | static void |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1186 | cfq_update_group_weight(struct cfq_group *cfqg) |
| 1187 | { |
| 1188 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
| 1189 | if (cfqg->needs_update) { |
| 1190 | cfqg->weight = cfqg->new_weight; |
| 1191 | cfqg->needs_update = false; |
| 1192 | } |
| 1193 | } |
| 1194 | |
| 1195 | static void |
| 1196 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1197 | { |
| 1198 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
| 1199 | |
| 1200 | cfq_update_group_weight(cfqg); |
| 1201 | __cfq_group_service_tree_add(st, cfqg); |
| 1202 | st->total_weight += cfqg->weight; |
| 1203 | } |
| 1204 | |
| 1205 | static void |
| 1206 | cfq_group_notify_queue_add(struct cfq_data *cfqd, struct cfq_group *cfqg) |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1207 | { |
| 1208 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 1209 | struct cfq_group *__cfqg; |
| 1210 | struct rb_node *n; |
| 1211 | |
| 1212 | cfqg->nr_cfqq++; |
Gui Jianfeng | 760701b | 2010-11-30 20:52:47 +0100 | [diff] [blame] | 1213 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1214 | return; |
| 1215 | |
| 1216 | /* |
| 1217 | * Currently put the group at the end. Later implement something |
| 1218 | * so that groups get lesser vtime based on their weights, so that |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 1219 | * if group does not loose all if it was not continuously backlogged. |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1220 | */ |
| 1221 | n = rb_last(&st->rb); |
| 1222 | if (n) { |
| 1223 | __cfqg = rb_entry_cfqg(n); |
| 1224 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; |
| 1225 | } else |
| 1226 | cfqg->vdisktime = st->min_vdisktime; |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1227 | cfq_group_service_tree_add(st, cfqg); |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1228 | } |
| 1229 | |
| 1230 | static void |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1231 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 1232 | { |
| 1233 | st->total_weight -= cfqg->weight; |
| 1234 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
| 1235 | cfq_rb_erase(&cfqg->rb_node, st); |
| 1236 | } |
| 1237 | |
| 1238 | static void |
| 1239 | cfq_group_notify_queue_del(struct cfq_data *cfqd, struct cfq_group *cfqg) |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1240 | { |
| 1241 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 1242 | |
| 1243 | BUG_ON(cfqg->nr_cfqq < 1); |
| 1244 | cfqg->nr_cfqq--; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 1245 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1246 | /* If there are other cfq queues under this group, don't delete it */ |
| 1247 | if (cfqg->nr_cfqq) |
| 1248 | return; |
| 1249 | |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 1250 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1251 | cfq_group_service_tree_del(st, cfqg); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1252 | cfqg->saved_workload_slice = 0; |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 1253 | cfq_blkiocg_update_dequeue_stats(cfqg_to_blkg(cfqg), |
| 1254 | &blkio_policy_cfq, 1); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1255 | } |
| 1256 | |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 1257 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
| 1258 | unsigned int *unaccounted_time) |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1259 | { |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 1260 | unsigned int slice_used; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1261 | |
| 1262 | /* |
| 1263 | * Queue got expired before even a single request completed or |
| 1264 | * got expired immediately after first request completion. |
| 1265 | */ |
| 1266 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { |
| 1267 | /* |
| 1268 | * Also charge the seek time incurred to the group, otherwise |
| 1269 | * if there are mutiple queues in the group, each can dispatch |
| 1270 | * a single request on seeky media and cause lots of seek time |
| 1271 | * and group will never know it. |
| 1272 | */ |
| 1273 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), |
| 1274 | 1); |
| 1275 | } else { |
| 1276 | slice_used = jiffies - cfqq->slice_start; |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 1277 | if (slice_used > cfqq->allocated_slice) { |
| 1278 | *unaccounted_time = slice_used - cfqq->allocated_slice; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 1279 | slice_used = cfqq->allocated_slice; |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 1280 | } |
| 1281 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) |
| 1282 | *unaccounted_time += cfqq->slice_start - |
| 1283 | cfqq->dispatch_start; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1284 | } |
| 1285 | |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1286 | return slice_used; |
| 1287 | } |
| 1288 | |
| 1289 | static void cfq_group_served(struct cfq_data *cfqd, struct cfq_group *cfqg, |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 1290 | struct cfq_queue *cfqq) |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1291 | { |
| 1292 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 1293 | unsigned int used_sl, charge, unaccounted_sl = 0; |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 1294 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
| 1295 | - cfqg->service_tree_idle.count; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1296 | |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 1297 | BUG_ON(nr_sync < 0); |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 1298 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 1299 | |
Vivek Goyal | 02b3508 | 2010-08-23 12:23:53 +0200 | [diff] [blame] | 1300 | if (iops_mode(cfqd)) |
| 1301 | charge = cfqq->slice_dispatch; |
| 1302 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) |
| 1303 | charge = cfqq->allocated_slice; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1304 | |
| 1305 | /* Can't update vdisktime while group is on service tree */ |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1306 | cfq_group_service_tree_del(st, cfqg); |
Vivek Goyal | 02b3508 | 2010-08-23 12:23:53 +0200 | [diff] [blame] | 1307 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1308 | /* If a new weight was requested, update now, off tree */ |
| 1309 | cfq_group_service_tree_add(st, cfqg); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1310 | |
| 1311 | /* This group is being expired. Save the context */ |
| 1312 | if (time_after(cfqd->workload_expires, jiffies)) { |
| 1313 | cfqg->saved_workload_slice = cfqd->workload_expires |
| 1314 | - jiffies; |
| 1315 | cfqg->saved_workload = cfqd->serving_type; |
| 1316 | cfqg->saved_serving_prio = cfqd->serving_prio; |
| 1317 | } else |
| 1318 | cfqg->saved_workload_slice = 0; |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 1319 | |
| 1320 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, |
| 1321 | st->min_vdisktime); |
Joe Perches | fd16d26 | 2011-06-13 10:42:49 +0200 | [diff] [blame] | 1322 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
| 1323 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", |
| 1324 | used_sl, cfqq->slice_dispatch, charge, |
| 1325 | iops_mode(cfqd), cfqq->nr_sectors); |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 1326 | cfq_blkiocg_update_timeslice_used(cfqg_to_blkg(cfqg), &blkio_policy_cfq, |
| 1327 | used_sl, unaccounted_sl); |
| 1328 | cfq_blkiocg_set_start_empty_time(cfqg_to_blkg(cfqg), &blkio_policy_cfq); |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1329 | } |
| 1330 | |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 1331 | /** |
| 1332 | * cfq_init_cfqg_base - initialize base part of a cfq_group |
| 1333 | * @cfqg: cfq_group to initialize |
| 1334 | * |
| 1335 | * Initialize the base part which is used whether %CONFIG_CFQ_GROUP_IOSCHED |
| 1336 | * is enabled or not. |
| 1337 | */ |
| 1338 | static void cfq_init_cfqg_base(struct cfq_group *cfqg) |
| 1339 | { |
| 1340 | struct cfq_rb_root *st; |
| 1341 | int i, j; |
| 1342 | |
| 1343 | for_each_cfqg_st(cfqg, i, j, st) |
| 1344 | *st = CFQ_RB_ROOT; |
| 1345 | RB_CLEAR_NODE(&cfqg->rb_node); |
| 1346 | |
| 1347 | cfqg->ttime.last_end_request = jiffies; |
| 1348 | } |
| 1349 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1350 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Tejun Heo | 60c2bc2 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 1351 | static void cfq_update_blkio_group_weight(struct blkio_group *blkg, |
Paul Bolle | 8aea454 | 2011-06-06 05:07:54 +0200 | [diff] [blame] | 1352 | unsigned int weight) |
Vivek Goyal | f8d461d | 2009-12-03 12:59:52 -0500 | [diff] [blame] | 1353 | { |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1354 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
| 1355 | |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1356 | cfqg->new_weight = weight; |
| 1357 | cfqg->needs_update = true; |
Vivek Goyal | f8d461d | 2009-12-03 12:59:52 -0500 | [diff] [blame] | 1358 | } |
| 1359 | |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1360 | static void cfq_init_blkio_group(struct blkio_group *blkg) |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1361 | { |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1362 | struct cfq_group *cfqg = blkg_to_cfqg(blkg); |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1363 | |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 1364 | cfq_init_cfqg_base(cfqg); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1365 | cfqg->weight = blkg->blkcg->weight; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1366 | } |
| 1367 | |
| 1368 | /* |
Vivek Goyal | 3e59cf9 | 2011-05-19 15:38:21 -0400 | [diff] [blame] | 1369 | * Search for the cfq group current task belongs to. request_queue lock must |
| 1370 | * be held. |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1371 | */ |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 1372 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
| 1373 | struct blkio_cgroup *blkcg) |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1374 | { |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1375 | struct request_queue *q = cfqd->queue; |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 1376 | struct cfq_group *cfqg = NULL; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1377 | |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 1378 | /* avoid lookup for the common case where there's no blkio cgroup */ |
| 1379 | if (blkcg == &blkio_root_cgroup) { |
| 1380 | cfqg = cfqd->root_group; |
| 1381 | } else { |
| 1382 | struct blkio_group *blkg; |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1383 | |
Tejun Heo | aaec55a | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 1384 | blkg = blkg_lookup_create(blkcg, q, false); |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 1385 | if (!IS_ERR(blkg)) |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1386 | cfqg = blkg_to_cfqg(blkg); |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1387 | } |
| 1388 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1389 | return cfqg; |
| 1390 | } |
| 1391 | |
| 1392 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) |
| 1393 | { |
| 1394 | /* Currently, all async queues are mapped to root group */ |
| 1395 | if (!cfq_cfqq_sync(cfqq)) |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 1396 | cfqg = cfqq->cfqd->root_group; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1397 | |
| 1398 | cfqq->cfqg = cfqg; |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1399 | /* cfqq reference on cfqg */ |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 1400 | cfqg_get(cfqg); |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1401 | } |
| 1402 | |
Tejun Heo | 60c2bc2 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 1403 | static u64 blkg_prfill_weight_device(struct seq_file *sf, |
| 1404 | struct blkg_policy_data *pd, int off) |
| 1405 | { |
| 1406 | if (!pd->conf.weight) |
| 1407 | return 0; |
| 1408 | return __blkg_prfill_u64(sf, pd, pd->conf.weight); |
| 1409 | } |
| 1410 | |
| 1411 | static int blkcg_print_weight_device(struct cgroup *cgrp, struct cftype *cft, |
| 1412 | struct seq_file *sf) |
| 1413 | { |
| 1414 | blkcg_print_blkgs(sf, cgroup_to_blkio_cgroup(cgrp), |
| 1415 | blkg_prfill_weight_device, BLKIO_POLICY_PROP, 0, |
| 1416 | false); |
| 1417 | return 0; |
| 1418 | } |
| 1419 | |
| 1420 | static int blkcg_print_weight(struct cgroup *cgrp, struct cftype *cft, |
| 1421 | struct seq_file *sf) |
| 1422 | { |
| 1423 | seq_printf(sf, "%u\n", cgroup_to_blkio_cgroup(cgrp)->weight); |
| 1424 | return 0; |
| 1425 | } |
| 1426 | |
| 1427 | static int blkcg_set_weight_device(struct cgroup *cgrp, struct cftype *cft, |
| 1428 | const char *buf) |
| 1429 | { |
| 1430 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
| 1431 | struct blkg_policy_data *pd; |
| 1432 | struct blkg_conf_ctx ctx; |
| 1433 | int ret; |
| 1434 | |
| 1435 | ret = blkg_conf_prep(blkcg, buf, &ctx); |
| 1436 | if (ret) |
| 1437 | return ret; |
| 1438 | |
| 1439 | ret = -EINVAL; |
| 1440 | pd = ctx.blkg->pd[BLKIO_POLICY_PROP]; |
| 1441 | if (pd && (!ctx.v || (ctx.v >= BLKIO_WEIGHT_MIN && |
| 1442 | ctx.v <= BLKIO_WEIGHT_MAX))) { |
| 1443 | pd->conf.weight = ctx.v; |
| 1444 | cfq_update_blkio_group_weight(ctx.blkg, ctx.v ?: blkcg->weight); |
| 1445 | ret = 0; |
| 1446 | } |
| 1447 | |
| 1448 | blkg_conf_finish(&ctx); |
| 1449 | return ret; |
| 1450 | } |
| 1451 | |
| 1452 | static int blkcg_set_weight(struct cgroup *cgrp, struct cftype *cft, u64 val) |
| 1453 | { |
| 1454 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
| 1455 | struct blkio_group *blkg; |
| 1456 | struct hlist_node *n; |
| 1457 | |
| 1458 | if (val < BLKIO_WEIGHT_MIN || val > BLKIO_WEIGHT_MAX) |
| 1459 | return -EINVAL; |
| 1460 | |
| 1461 | spin_lock_irq(&blkcg->lock); |
| 1462 | blkcg->weight = (unsigned int)val; |
| 1463 | |
| 1464 | hlist_for_each_entry(blkg, n, &blkcg->blkg_list, blkcg_node) { |
| 1465 | struct blkg_policy_data *pd = blkg->pd[BLKIO_POLICY_PROP]; |
| 1466 | |
| 1467 | if (pd && !pd->conf.weight) |
| 1468 | cfq_update_blkio_group_weight(blkg, blkcg->weight); |
| 1469 | } |
| 1470 | |
| 1471 | spin_unlock_irq(&blkcg->lock); |
| 1472 | return 0; |
| 1473 | } |
| 1474 | |
| 1475 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 1476 | static u64 blkg_prfill_avg_queue_size(struct seq_file *sf, |
| 1477 | struct blkg_policy_data *pd, int off) |
| 1478 | { |
| 1479 | u64 samples = blkg_stat_read(&pd->stats.avg_queue_size_samples); |
| 1480 | u64 v = 0; |
| 1481 | |
| 1482 | if (samples) { |
| 1483 | v = blkg_stat_read(&pd->stats.avg_queue_size_sum); |
| 1484 | do_div(v, samples); |
| 1485 | } |
| 1486 | __blkg_prfill_u64(sf, pd, v); |
| 1487 | return 0; |
| 1488 | } |
| 1489 | |
| 1490 | /* print avg_queue_size */ |
| 1491 | static int blkcg_print_avg_queue_size(struct cgroup *cgrp, struct cftype *cft, |
| 1492 | struct seq_file *sf) |
| 1493 | { |
| 1494 | struct blkio_cgroup *blkcg = cgroup_to_blkio_cgroup(cgrp); |
| 1495 | |
| 1496 | blkcg_print_blkgs(sf, blkcg, blkg_prfill_avg_queue_size, |
| 1497 | BLKIO_POLICY_PROP, 0, false); |
| 1498 | return 0; |
| 1499 | } |
| 1500 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ |
| 1501 | |
| 1502 | static struct cftype cfq_blkcg_files[] = { |
| 1503 | { |
| 1504 | .name = "weight_device", |
| 1505 | .read_seq_string = blkcg_print_weight_device, |
| 1506 | .write_string = blkcg_set_weight_device, |
| 1507 | .max_write_len = 256, |
| 1508 | }, |
| 1509 | { |
| 1510 | .name = "weight", |
| 1511 | .read_seq_string = blkcg_print_weight, |
| 1512 | .write_u64 = blkcg_set_weight, |
| 1513 | }, |
| 1514 | { |
| 1515 | .name = "time", |
| 1516 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1517 | offsetof(struct blkio_group_stats, time)), |
| 1518 | .read_seq_string = blkcg_print_stat, |
| 1519 | }, |
| 1520 | { |
| 1521 | .name = "sectors", |
| 1522 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1523 | offsetof(struct blkio_group_stats_cpu, sectors)), |
| 1524 | .read_seq_string = blkcg_print_cpu_stat, |
| 1525 | }, |
| 1526 | { |
| 1527 | .name = "io_service_bytes", |
| 1528 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1529 | offsetof(struct blkio_group_stats_cpu, service_bytes)), |
| 1530 | .read_seq_string = blkcg_print_cpu_rwstat, |
| 1531 | }, |
| 1532 | { |
| 1533 | .name = "io_serviced", |
| 1534 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1535 | offsetof(struct blkio_group_stats_cpu, serviced)), |
| 1536 | .read_seq_string = blkcg_print_cpu_rwstat, |
| 1537 | }, |
| 1538 | { |
| 1539 | .name = "io_service_time", |
| 1540 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1541 | offsetof(struct blkio_group_stats, service_time)), |
| 1542 | .read_seq_string = blkcg_print_rwstat, |
| 1543 | }, |
| 1544 | { |
| 1545 | .name = "io_wait_time", |
| 1546 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1547 | offsetof(struct blkio_group_stats, wait_time)), |
| 1548 | .read_seq_string = blkcg_print_rwstat, |
| 1549 | }, |
| 1550 | { |
| 1551 | .name = "io_merged", |
| 1552 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1553 | offsetof(struct blkio_group_stats, merged)), |
| 1554 | .read_seq_string = blkcg_print_rwstat, |
| 1555 | }, |
| 1556 | { |
| 1557 | .name = "io_queued", |
| 1558 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1559 | offsetof(struct blkio_group_stats, queued)), |
| 1560 | .read_seq_string = blkcg_print_rwstat, |
| 1561 | }, |
| 1562 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
| 1563 | { |
| 1564 | .name = "avg_queue_size", |
| 1565 | .read_seq_string = blkcg_print_avg_queue_size, |
| 1566 | }, |
| 1567 | { |
| 1568 | .name = "group_wait_time", |
| 1569 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1570 | offsetof(struct blkio_group_stats, group_wait_time)), |
| 1571 | .read_seq_string = blkcg_print_stat, |
| 1572 | }, |
| 1573 | { |
| 1574 | .name = "idle_time", |
| 1575 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1576 | offsetof(struct blkio_group_stats, idle_time)), |
| 1577 | .read_seq_string = blkcg_print_stat, |
| 1578 | }, |
| 1579 | { |
| 1580 | .name = "empty_time", |
| 1581 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1582 | offsetof(struct blkio_group_stats, empty_time)), |
| 1583 | .read_seq_string = blkcg_print_stat, |
| 1584 | }, |
| 1585 | { |
| 1586 | .name = "dequeue", |
| 1587 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1588 | offsetof(struct blkio_group_stats, dequeue)), |
| 1589 | .read_seq_string = blkcg_print_stat, |
| 1590 | }, |
| 1591 | { |
| 1592 | .name = "unaccounted_time", |
| 1593 | .private = BLKCG_STAT_PRIV(BLKIO_POLICY_PROP, |
| 1594 | offsetof(struct blkio_group_stats, unaccounted_time)), |
| 1595 | .read_seq_string = blkcg_print_stat, |
| 1596 | }, |
| 1597 | #endif /* CONFIG_DEBUG_BLK_CGROUP */ |
| 1598 | { } /* terminate */ |
| 1599 | }; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1600 | #else /* GROUP_IOSCHED */ |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 1601 | static struct cfq_group *cfq_lookup_create_cfqg(struct cfq_data *cfqd, |
| 1602 | struct blkio_cgroup *blkcg) |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1603 | { |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 1604 | return cfqd->root_group; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1605 | } |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 1606 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1607 | static inline void |
| 1608 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { |
| 1609 | cfqq->cfqg = cfqg; |
| 1610 | } |
| 1611 | |
| 1612 | #endif /* GROUP_IOSCHED */ |
| 1613 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1614 | /* |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1615 | * The cfqd->service_trees holds all pending cfq_queue's that have |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1616 | * requests waiting to be processed. It is sorted in the order that |
| 1617 | * we will service the queues. |
| 1618 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1619 | static void cfq_service_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 1620 | bool add_front) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1621 | { |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1622 | struct rb_node **p, *parent; |
| 1623 | struct cfq_queue *__cfqq; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1624 | unsigned long rb_key; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1625 | struct cfq_rb_root *service_tree; |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1626 | int left; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1627 | int new_cfqq = 1; |
Vivek Goyal | ae30c28 | 2009-12-03 12:59:55 -0500 | [diff] [blame] | 1628 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 1629 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 1630 | cfqq_type(cfqq)); |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1631 | if (cfq_class_idle(cfqq)) { |
| 1632 | rb_key = CFQ_IDLE_DELAY; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1633 | parent = rb_last(&service_tree->rb); |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1634 | if (parent && parent != &cfqq->rb_node) { |
| 1635 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
| 1636 | rb_key += __cfqq->rb_key; |
| 1637 | } else |
| 1638 | rb_key += jiffies; |
| 1639 | } else if (!add_front) { |
Jens Axboe | b9c8946 | 2009-10-06 20:53:44 +0200 | [diff] [blame] | 1640 | /* |
| 1641 | * Get our rb key offset. Subtract any residual slice |
| 1642 | * value carried from last service. A negative resid |
| 1643 | * count indicates slice overrun, and this should position |
| 1644 | * the next service time further away in the tree. |
| 1645 | */ |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1646 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
Jens Axboe | b9c8946 | 2009-10-06 20:53:44 +0200 | [diff] [blame] | 1647 | rb_key -= cfqq->slice_resid; |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1648 | cfqq->slice_resid = 0; |
Corrado Zoccolo | 48e025e | 2009-10-05 08:49:23 +0200 | [diff] [blame] | 1649 | } else { |
| 1650 | rb_key = -HZ; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1651 | __cfqq = cfq_rb_first(service_tree); |
Corrado Zoccolo | 48e025e | 2009-10-05 08:49:23 +0200 | [diff] [blame] | 1652 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
| 1653 | } |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1654 | |
| 1655 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1656 | new_cfqq = 0; |
Jens Axboe | 99f9628 | 2007-02-05 11:56:25 +0100 | [diff] [blame] | 1657 | /* |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1658 | * same position, nothing more to do |
Jens Axboe | 99f9628 | 2007-02-05 11:56:25 +0100 | [diff] [blame] | 1659 | */ |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1660 | if (rb_key == cfqq->rb_key && |
| 1661 | cfqq->service_tree == service_tree) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1662 | return; |
Jens Axboe | 53b03744 | 2006-07-28 09:48:51 +0200 | [diff] [blame] | 1663 | |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1664 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
| 1665 | cfqq->service_tree = NULL; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1666 | } |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1667 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1668 | left = 1; |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1669 | parent = NULL; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1670 | cfqq->service_tree = service_tree; |
| 1671 | p = &service_tree->rb.rb_node; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1672 | while (*p) { |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1673 | struct rb_node **n; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1674 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1675 | parent = *p; |
| 1676 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
| 1677 | |
Jens Axboe | 0c534e0 | 2007-04-18 20:01:57 +0200 | [diff] [blame] | 1678 | /* |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1679 | * sort by key, that represents service time. |
Jens Axboe | 0c534e0 | 2007-04-18 20:01:57 +0200 | [diff] [blame] | 1680 | */ |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1681 | if (time_before(rb_key, __cfqq->rb_key)) |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1682 | n = &(*p)->rb_left; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1683 | else { |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1684 | n = &(*p)->rb_right; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1685 | left = 0; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1686 | } |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1687 | |
| 1688 | p = n; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1689 | } |
| 1690 | |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1691 | if (left) |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1692 | service_tree->left = &cfqq->rb_node; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1693 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1694 | cfqq->rb_key = rb_key; |
| 1695 | rb_link_node(&cfqq->rb_node, parent, p); |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1696 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
| 1697 | service_tree->count++; |
Namhyung Kim | 20359f2 | 2011-05-24 10:23:22 +0200 | [diff] [blame] | 1698 | if (add_front || !new_cfqq) |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1699 | return; |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1700 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1701 | } |
| 1702 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1703 | static struct cfq_queue * |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1704 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
| 1705 | sector_t sector, struct rb_node **ret_parent, |
| 1706 | struct rb_node ***rb_link) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1707 | { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1708 | struct rb_node **p, *parent; |
| 1709 | struct cfq_queue *cfqq = NULL; |
| 1710 | |
| 1711 | parent = NULL; |
| 1712 | p = &root->rb_node; |
| 1713 | while (*p) { |
| 1714 | struct rb_node **n; |
| 1715 | |
| 1716 | parent = *p; |
| 1717 | cfqq = rb_entry(parent, struct cfq_queue, p_node); |
| 1718 | |
| 1719 | /* |
| 1720 | * Sort strictly based on sector. Smallest to the left, |
| 1721 | * largest to the right. |
| 1722 | */ |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 1723 | if (sector > blk_rq_pos(cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1724 | n = &(*p)->rb_right; |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 1725 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1726 | n = &(*p)->rb_left; |
| 1727 | else |
| 1728 | break; |
| 1729 | p = n; |
Jens Axboe | 3ac6c9f | 2009-04-23 12:14:56 +0200 | [diff] [blame] | 1730 | cfqq = NULL; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1731 | } |
| 1732 | |
| 1733 | *ret_parent = parent; |
| 1734 | if (rb_link) |
| 1735 | *rb_link = p; |
Jens Axboe | 3ac6c9f | 2009-04-23 12:14:56 +0200 | [diff] [blame] | 1736 | return cfqq; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1737 | } |
| 1738 | |
| 1739 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 1740 | { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1741 | struct rb_node **p, *parent; |
| 1742 | struct cfq_queue *__cfqq; |
| 1743 | |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1744 | if (cfqq->p_root) { |
| 1745 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 1746 | cfqq->p_root = NULL; |
| 1747 | } |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1748 | |
| 1749 | if (cfq_class_idle(cfqq)) |
| 1750 | return; |
| 1751 | if (!cfqq->next_rq) |
| 1752 | return; |
| 1753 | |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1754 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 1755 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
| 1756 | blk_rq_pos(cfqq->next_rq), &parent, &p); |
Jens Axboe | 3ac6c9f | 2009-04-23 12:14:56 +0200 | [diff] [blame] | 1757 | if (!__cfqq) { |
| 1758 | rb_link_node(&cfqq->p_node, parent, p); |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1759 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
| 1760 | } else |
| 1761 | cfqq->p_root = NULL; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1762 | } |
| 1763 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1764 | /* |
| 1765 | * Update cfqq's position in the service tree. |
| 1766 | */ |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1767 | static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1768 | { |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1769 | /* |
| 1770 | * Resorting requires the cfqq to be on the RR list already. |
| 1771 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1772 | if (cfq_cfqq_on_rr(cfqq)) { |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1773 | cfq_service_tree_add(cfqd, cfqq, 0); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1774 | cfq_prio_tree_add(cfqd, cfqq); |
| 1775 | } |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1776 | } |
| 1777 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1778 | /* |
| 1779 | * add to busy list of queues for service, trying to be fair in ordering |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1780 | * the pending list according to last request service |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1781 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1782 | static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1783 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1784 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 1785 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
| 1786 | cfq_mark_cfqq_on_rr(cfqq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1787 | cfqd->busy_queues++; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 1788 | if (cfq_cfqq_sync(cfqq)) |
| 1789 | cfqd->busy_sync_queues++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1790 | |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1791 | cfq_resort_rr_list(cfqd, cfqq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1792 | } |
| 1793 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1794 | /* |
| 1795 | * Called when the cfqq no longer has requests pending, remove it from |
| 1796 | * the service tree. |
| 1797 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1798 | static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1799 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1800 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 1801 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
| 1802 | cfq_clear_cfqq_on_rr(cfqq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1803 | |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1804 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
| 1805 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
| 1806 | cfqq->service_tree = NULL; |
| 1807 | } |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1808 | if (cfqq->p_root) { |
| 1809 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 1810 | cfqq->p_root = NULL; |
| 1811 | } |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1812 | |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1813 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1814 | BUG_ON(!cfqd->busy_queues); |
| 1815 | cfqd->busy_queues--; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 1816 | if (cfq_cfqq_sync(cfqq)) |
| 1817 | cfqd->busy_sync_queues--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1818 | } |
| 1819 | |
| 1820 | /* |
| 1821 | * rb tree support functions |
| 1822 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1823 | static void cfq_del_rq_rb(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1824 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1825 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1826 | const int sync = rq_is_sync(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1827 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1828 | BUG_ON(!cfqq->queued[sync]); |
| 1829 | cfqq->queued[sync]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1830 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1831 | elv_rb_del(&cfqq->sort_list, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1832 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1833 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
| 1834 | /* |
| 1835 | * Queue will be deleted from service tree when we actually |
| 1836 | * expire it later. Right now just remove it from prio tree |
| 1837 | * as it is empty. |
| 1838 | */ |
| 1839 | if (cfqq->p_root) { |
| 1840 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 1841 | cfqq->p_root = NULL; |
| 1842 | } |
| 1843 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1844 | } |
| 1845 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1846 | static void cfq_add_rq_rb(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1847 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1848 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1849 | struct cfq_data *cfqd = cfqq->cfqd; |
Jeff Moyer | 796d511 | 2011-06-02 21:19:05 +0200 | [diff] [blame] | 1850 | struct request *prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1851 | |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 1852 | cfqq->queued[rq_is_sync(rq)]++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1853 | |
Jeff Moyer | 796d511 | 2011-06-02 21:19:05 +0200 | [diff] [blame] | 1854 | elv_rb_add(&cfqq->sort_list, rq); |
Jens Axboe | 5fccbf6 | 2006-10-31 14:21:55 +0100 | [diff] [blame] | 1855 | |
| 1856 | if (!cfq_cfqq_on_rr(cfqq)) |
| 1857 | cfq_add_cfqq_rr(cfqd, cfqq); |
Jens Axboe | 5044eed | 2007-04-25 11:53:48 +0200 | [diff] [blame] | 1858 | |
| 1859 | /* |
| 1860 | * check if this request is a better next-serve candidate |
| 1861 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1862 | prev = cfqq->next_rq; |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 1863 | cfqq->next_rq = cfq_choose_req(cfqd, cfqq->next_rq, rq, cfqd->last_position); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1864 | |
| 1865 | /* |
| 1866 | * adjust priority tree position, if ->next_rq changes |
| 1867 | */ |
| 1868 | if (prev != cfqq->next_rq) |
| 1869 | cfq_prio_tree_add(cfqd, cfqq); |
| 1870 | |
Jens Axboe | 5044eed | 2007-04-25 11:53:48 +0200 | [diff] [blame] | 1871 | BUG_ON(!cfqq->next_rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1872 | } |
| 1873 | |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1874 | static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1875 | { |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 1876 | elv_rb_del(&cfqq->sort_list, rq); |
| 1877 | cfqq->queued[rq_is_sync(rq)]--; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1878 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 1879 | &blkio_policy_cfq, rq_data_dir(rq), |
| 1880 | rq_is_sync(rq)); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1881 | cfq_add_rq_rb(rq); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1882 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 1883 | &blkio_policy_cfq, |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1884 | cfqg_to_blkg(cfqq->cfqd->serving_group), |
| 1885 | rq_data_dir(rq), rq_is_sync(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1886 | } |
| 1887 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1888 | static struct request * |
| 1889 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1890 | { |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1891 | struct task_struct *tsk = current; |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 1892 | struct cfq_io_cq *cic; |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1893 | struct cfq_queue *cfqq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1894 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 1895 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 1896 | if (!cic) |
| 1897 | return NULL; |
| 1898 | |
| 1899 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 1900 | if (cfqq) { |
| 1901 | sector_t sector = bio->bi_sector + bio_sectors(bio); |
| 1902 | |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1903 | return elv_rb_find(&cfqq->sort_list, sector); |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 1904 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1905 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1906 | return NULL; |
| 1907 | } |
| 1908 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1909 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1910 | { |
| 1911 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 1912 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1913 | cfqd->rq_in_driver++; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1914 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1915 | cfqd->rq_in_driver); |
Jens Axboe | 25776e3 | 2006-06-01 10:12:26 +0200 | [diff] [blame] | 1916 | |
Tejun Heo | 5b93629 | 2009-05-07 22:24:38 +0900 | [diff] [blame] | 1917 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1918 | } |
| 1919 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1920 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1921 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1922 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1923 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1924 | WARN_ON(!cfqd->rq_in_driver); |
| 1925 | cfqd->rq_in_driver--; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1926 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1927 | cfqd->rq_in_driver); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1928 | } |
| 1929 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1930 | static void cfq_remove_request(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1931 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1932 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1933 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1934 | if (cfqq->next_rq == rq) |
| 1935 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1936 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1937 | list_del_init(&rq->queuelist); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1938 | cfq_del_rq_rb(rq); |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 1939 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 1940 | cfqq->cfqd->rq_queued--; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1941 | cfq_blkiocg_update_io_remove_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 1942 | &blkio_policy_cfq, rq_data_dir(rq), |
| 1943 | rq_is_sync(rq)); |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 1944 | if (rq->cmd_flags & REQ_PRIO) { |
| 1945 | WARN_ON(!cfqq->prio_pending); |
| 1946 | cfqq->prio_pending--; |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 1947 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1948 | } |
| 1949 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1950 | static int cfq_merge(struct request_queue *q, struct request **req, |
| 1951 | struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1952 | { |
| 1953 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 1954 | struct request *__rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1955 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1956 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1957 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
Jens Axboe | 9817064 | 2006-07-28 09:23:08 +0200 | [diff] [blame] | 1958 | *req = __rq; |
| 1959 | return ELEVATOR_FRONT_MERGE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1960 | } |
| 1961 | |
| 1962 | return ELEVATOR_NO_MERGE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1963 | } |
| 1964 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1965 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1966 | int type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | { |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1968 | if (type == ELEVATOR_FRONT_MERGE) { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1969 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1970 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1971 | cfq_reposition_rq_rb(cfqq, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1972 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1973 | } |
| 1974 | |
Divyesh Shah | 812d402 | 2010-04-08 21:14:23 -0700 | [diff] [blame] | 1975 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
| 1976 | struct bio *bio) |
| 1977 | { |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 1978 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(req)), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 1979 | &blkio_policy_cfq, bio_data_dir(bio), |
| 1980 | cfq_bio_sync(bio)); |
Divyesh Shah | 812d402 | 2010-04-08 21:14:23 -0700 | [diff] [blame] | 1981 | } |
| 1982 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1983 | static void |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1984 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1985 | struct request *next) |
| 1986 | { |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 1987 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Shaohua Li | 4a0b75c | 2011-12-16 14:00:22 +0100 | [diff] [blame] | 1988 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 1989 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1990 | /* |
| 1991 | * reposition in fifo if next is older than rq |
| 1992 | */ |
| 1993 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 1994 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1995 | list_move(&rq->queuelist, &next->queuelist); |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 1996 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
| 1997 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1998 | |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 1999 | if (cfqq->next_rq == next) |
| 2000 | cfqq->next_rq = rq; |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 2001 | cfq_remove_request(next); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 2002 | cfq_blkiocg_update_io_merged_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 2003 | &blkio_policy_cfq, rq_data_dir(next), |
| 2004 | rq_is_sync(next)); |
Shaohua Li | 4a0b75c | 2011-12-16 14:00:22 +0100 | [diff] [blame] | 2005 | |
| 2006 | cfqq = RQ_CFQQ(next); |
| 2007 | /* |
| 2008 | * all requests of this queue are merged to other queues, delete it |
| 2009 | * from the service tree. If it's the active_queue, |
| 2010 | * cfq_dispatch_requests() will choose to expire it or do idle |
| 2011 | */ |
| 2012 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list) && |
| 2013 | cfqq != cfqd->active_queue) |
| 2014 | cfq_del_cfqq_rr(cfqd, cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2015 | } |
| 2016 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 2017 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 2018 | struct bio *bio) |
| 2019 | { |
| 2020 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 2021 | struct cfq_io_cq *cic; |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 2022 | struct cfq_queue *cfqq; |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 2023 | |
| 2024 | /* |
Jens Axboe | ec8acb6 | 2007-01-02 18:32:11 +0100 | [diff] [blame] | 2025 | * Disallow merge of a sync bio into an async request. |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 2026 | */ |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 2027 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 2028 | return false; |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 2029 | |
| 2030 | /* |
Tejun Heo | f1a4f4d | 2011-12-14 00:33:39 +0100 | [diff] [blame] | 2031 | * Lookup the cfqq that this bio will be queued with and allow |
Tejun Heo | 07c2bd3 | 2012-02-08 09:19:42 +0100 | [diff] [blame] | 2032 | * merge only if rq is queued there. |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 2033 | */ |
Tejun Heo | 07c2bd3 | 2012-02-08 09:19:42 +0100 | [diff] [blame] | 2034 | cic = cfq_cic_lookup(cfqd, current->io_context); |
| 2035 | if (!cic) |
| 2036 | return false; |
Jens Axboe | 719d340 | 2006-12-22 09:38:53 +0100 | [diff] [blame] | 2037 | |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 2038 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 2039 | return cfqq == RQ_CFQQ(rq); |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 2040 | } |
| 2041 | |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 2042 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2043 | { |
| 2044 | del_timer(&cfqd->idle_slice_timer); |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 2045 | cfq_blkiocg_update_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
| 2046 | &blkio_policy_cfq); |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 2047 | } |
| 2048 | |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 2049 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
| 2050 | struct cfq_queue *cfqq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2051 | { |
| 2052 | if (cfqq) { |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 2053 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
| 2054 | cfqd->serving_prio, cfqd->serving_type); |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 2055 | cfq_blkiocg_update_avg_queue_size_stats(cfqg_to_blkg(cfqq->cfqg), |
| 2056 | &blkio_policy_cfq); |
Justin TerAvest | 62a37f6 | 2011-03-23 08:25:44 +0100 | [diff] [blame] | 2057 | cfqq->slice_start = 0; |
| 2058 | cfqq->dispatch_start = jiffies; |
| 2059 | cfqq->allocated_slice = 0; |
| 2060 | cfqq->slice_end = 0; |
| 2061 | cfqq->slice_dispatch = 0; |
| 2062 | cfqq->nr_sectors = 0; |
| 2063 | |
| 2064 | cfq_clear_cfqq_wait_request(cfqq); |
| 2065 | cfq_clear_cfqq_must_dispatch(cfqq); |
| 2066 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
| 2067 | cfq_clear_cfqq_fifo_expire(cfqq); |
| 2068 | cfq_mark_cfqq_slice_new(cfqq); |
| 2069 | |
| 2070 | cfq_del_timer(cfqd, cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2071 | } |
| 2072 | |
| 2073 | cfqd->active_queue = cfqq; |
| 2074 | } |
| 2075 | |
| 2076 | /* |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2077 | * current cfqq expired its slice (or was too idle), select new one |
| 2078 | */ |
| 2079 | static void |
| 2080 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2081 | bool timed_out) |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2082 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2083 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
| 2084 | |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2085 | if (cfq_cfqq_wait_request(cfqq)) |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 2086 | cfq_del_timer(cfqd, cfqq); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2087 | |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2088 | cfq_clear_cfqq_wait_request(cfqq); |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 2089 | cfq_clear_cfqq_wait_busy(cfqq); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2090 | |
| 2091 | /* |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 2092 | * If this cfqq is shared between multiple processes, check to |
| 2093 | * make sure that those processes are still issuing I/Os within |
| 2094 | * the mean seek distance. If not, it may be time to break the |
| 2095 | * queues apart again. |
| 2096 | */ |
| 2097 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) |
| 2098 | cfq_mark_cfqq_split_coop(cfqq); |
| 2099 | |
| 2100 | /* |
Jens Axboe | 6084cdd | 2007-04-23 08:25:00 +0200 | [diff] [blame] | 2101 | * store what was left of this slice, if the queue idled/timed out |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2102 | */ |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 2103 | if (timed_out) { |
| 2104 | if (cfq_cfqq_slice_new(cfqq)) |
Vivek Goyal | ba5bd52 | 2011-01-19 08:25:02 -0700 | [diff] [blame] | 2105 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 2106 | else |
| 2107 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2108 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
| 2109 | } |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2110 | |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2111 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 2112 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2113 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 2114 | cfq_del_cfqq_rr(cfqd, cfqq); |
| 2115 | |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 2116 | cfq_resort_rr_list(cfqd, cfqq); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2117 | |
| 2118 | if (cfqq == cfqd->active_queue) |
| 2119 | cfqd->active_queue = NULL; |
| 2120 | |
| 2121 | if (cfqd->active_cic) { |
Tejun Heo | 11a3122 | 2012-02-07 07:51:30 +0100 | [diff] [blame] | 2122 | put_io_context(cfqd->active_cic->icq.ioc); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2123 | cfqd->active_cic = NULL; |
| 2124 | } |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2125 | } |
| 2126 | |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2127 | static inline void cfq_slice_expired(struct cfq_data *cfqd, bool timed_out) |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2128 | { |
| 2129 | struct cfq_queue *cfqq = cfqd->active_queue; |
| 2130 | |
| 2131 | if (cfqq) |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2132 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2133 | } |
| 2134 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2135 | /* |
| 2136 | * Get next queue for service. Unless we have a queue preemption, |
| 2137 | * we'll simply select the first cfqq in the service tree. |
| 2138 | */ |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2139 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2140 | { |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 2141 | struct cfq_rb_root *service_tree = |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2142 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2143 | cfqd->serving_type); |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 2144 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2145 | if (!cfqd->rq_queued) |
| 2146 | return NULL; |
| 2147 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2148 | /* There is nothing to dispatch */ |
| 2149 | if (!service_tree) |
| 2150 | return NULL; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 2151 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
| 2152 | return NULL; |
| 2153 | return cfq_rb_first(service_tree); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2154 | } |
| 2155 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2156 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
| 2157 | { |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 2158 | struct cfq_group *cfqg; |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2159 | struct cfq_queue *cfqq; |
| 2160 | int i, j; |
| 2161 | struct cfq_rb_root *st; |
| 2162 | |
| 2163 | if (!cfqd->rq_queued) |
| 2164 | return NULL; |
| 2165 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 2166 | cfqg = cfq_get_next_cfqg(cfqd); |
| 2167 | if (!cfqg) |
| 2168 | return NULL; |
| 2169 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2170 | for_each_cfqg_st(cfqg, i, j, st) |
| 2171 | if ((cfqq = cfq_rb_first(st)) != NULL) |
| 2172 | return cfqq; |
| 2173 | return NULL; |
| 2174 | } |
| 2175 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2176 | /* |
| 2177 | * Get and set a new active queue for service. |
| 2178 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2179 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
| 2180 | struct cfq_queue *cfqq) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2181 | { |
Jens Axboe | e00ef79 | 2009-11-04 08:54:55 +0100 | [diff] [blame] | 2182 | if (!cfqq) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2183 | cfqq = cfq_get_next_queue(cfqd); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2184 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2185 | __cfq_set_active_queue(cfqd, cfqq); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2186 | return cfqq; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2187 | } |
| 2188 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2189 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
| 2190 | struct request *rq) |
| 2191 | { |
Tejun Heo | 83096eb | 2009-05-07 22:24:39 +0900 | [diff] [blame] | 2192 | if (blk_rq_pos(rq) >= cfqd->last_position) |
| 2193 | return blk_rq_pos(rq) - cfqd->last_position; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2194 | else |
Tejun Heo | 83096eb | 2009-05-07 22:24:39 +0900 | [diff] [blame] | 2195 | return cfqd->last_position - blk_rq_pos(rq); |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2196 | } |
| 2197 | |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 2198 | static inline int cfq_rq_close(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 2199 | struct request *rq) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2200 | { |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 2201 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2202 | } |
| 2203 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2204 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
| 2205 | struct cfq_queue *cur_cfqq) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2206 | { |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 2207 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2208 | struct rb_node *parent, *node; |
| 2209 | struct cfq_queue *__cfqq; |
| 2210 | sector_t sector = cfqd->last_position; |
| 2211 | |
| 2212 | if (RB_EMPTY_ROOT(root)) |
| 2213 | return NULL; |
| 2214 | |
| 2215 | /* |
| 2216 | * First, if we find a request starting at the end of the last |
| 2217 | * request, choose it. |
| 2218 | */ |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 2219 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2220 | if (__cfqq) |
| 2221 | return __cfqq; |
| 2222 | |
| 2223 | /* |
| 2224 | * If the exact sector wasn't found, the parent of the NULL leaf |
| 2225 | * will contain the closest sector. |
| 2226 | */ |
| 2227 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 2228 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2229 | return __cfqq; |
| 2230 | |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 2231 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2232 | node = rb_next(&__cfqq->p_node); |
| 2233 | else |
| 2234 | node = rb_prev(&__cfqq->p_node); |
| 2235 | if (!node) |
| 2236 | return NULL; |
| 2237 | |
| 2238 | __cfqq = rb_entry(node, struct cfq_queue, p_node); |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 2239 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2240 | return __cfqq; |
| 2241 | |
| 2242 | return NULL; |
| 2243 | } |
| 2244 | |
| 2245 | /* |
| 2246 | * cfqd - obvious |
| 2247 | * cur_cfqq - passed in so that we don't decide that the current queue is |
| 2248 | * closely cooperating with itself. |
| 2249 | * |
| 2250 | * So, basically we're assuming that that cur_cfqq has dispatched at least |
| 2251 | * one request, and that cfqd->last_position reflects a position on the disk |
| 2252 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid |
| 2253 | * assumption. |
| 2254 | */ |
| 2255 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 2256 | struct cfq_queue *cur_cfqq) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2257 | { |
| 2258 | struct cfq_queue *cfqq; |
| 2259 | |
Divyesh Shah | 39c01b2 | 2010-03-25 15:45:57 +0100 | [diff] [blame] | 2260 | if (cfq_class_idle(cur_cfqq)) |
| 2261 | return NULL; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2262 | if (!cfq_cfqq_sync(cur_cfqq)) |
| 2263 | return NULL; |
| 2264 | if (CFQQ_SEEKY(cur_cfqq)) |
| 2265 | return NULL; |
| 2266 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2267 | /* |
Gui Jianfeng | b9d8f4c | 2009-12-08 08:54:17 +0100 | [diff] [blame] | 2268 | * Don't search priority tree if it's the only queue in the group. |
| 2269 | */ |
| 2270 | if (cur_cfqq->cfqg->nr_cfqq == 1) |
| 2271 | return NULL; |
| 2272 | |
| 2273 | /* |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2274 | * We should notice if some of the queues are cooperating, eg |
| 2275 | * working closely on the same area of the disk. In that case, |
| 2276 | * we can group them together and don't waste time idling. |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2277 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2278 | cfqq = cfqq_close(cfqd, cur_cfqq); |
| 2279 | if (!cfqq) |
| 2280 | return NULL; |
| 2281 | |
Vivek Goyal | 8682e1f | 2009-12-03 12:59:50 -0500 | [diff] [blame] | 2282 | /* If new queue belongs to different cfq_group, don't choose it */ |
| 2283 | if (cur_cfqq->cfqg != cfqq->cfqg) |
| 2284 | return NULL; |
| 2285 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2286 | /* |
| 2287 | * It only makes sense to merge sync queues. |
| 2288 | */ |
| 2289 | if (!cfq_cfqq_sync(cfqq)) |
| 2290 | return NULL; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2291 | if (CFQQ_SEEKY(cfqq)) |
| 2292 | return NULL; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2293 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 2294 | /* |
| 2295 | * Do not merge queues of different priority classes |
| 2296 | */ |
| 2297 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) |
| 2298 | return NULL; |
| 2299 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2300 | return cfqq; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2301 | } |
| 2302 | |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 2303 | /* |
| 2304 | * Determine whether we should enforce idle window for this queue. |
| 2305 | */ |
| 2306 | |
| 2307 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2308 | { |
| 2309 | enum wl_prio_t prio = cfqq_prio(cfqq); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2310 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 2311 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2312 | BUG_ON(!service_tree); |
| 2313 | BUG_ON(!service_tree->count); |
| 2314 | |
Vivek Goyal | b6508c1 | 2010-08-23 12:23:33 +0200 | [diff] [blame] | 2315 | if (!cfqd->cfq_slice_idle) |
| 2316 | return false; |
| 2317 | |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 2318 | /* We never do for idle class queues. */ |
| 2319 | if (prio == IDLE_WORKLOAD) |
| 2320 | return false; |
| 2321 | |
| 2322 | /* We do for queues that were marked with idle window flag. */ |
Shaohua Li | 3c764b7 | 2009-12-04 13:12:06 +0100 | [diff] [blame] | 2323 | if (cfq_cfqq_idle_window(cfqq) && |
| 2324 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 2325 | return true; |
| 2326 | |
| 2327 | /* |
| 2328 | * Otherwise, we do only if they are the last ones |
| 2329 | * in their service tree. |
| 2330 | */ |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 2331 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
| 2332 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2333 | return true; |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 2334 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
| 2335 | service_tree->count); |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2336 | return false; |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 2337 | } |
| 2338 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2339 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2340 | { |
Jens Axboe | 1792669 | 2007-01-19 11:59:30 +1100 | [diff] [blame] | 2341 | struct cfq_queue *cfqq = cfqd->active_queue; |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 2342 | struct cfq_io_cq *cic; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2343 | unsigned long sl, group_idle = 0; |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2344 | |
Jens Axboe | a68bbdd | 2008-09-24 13:03:33 +0200 | [diff] [blame] | 2345 | /* |
Jens Axboe | f7d7b7a | 2008-09-25 11:37:50 +0200 | [diff] [blame] | 2346 | * SSD device without seek penalty, disable idling. But only do so |
| 2347 | * for devices that support queuing, otherwise we still have a problem |
| 2348 | * with sync vs async workloads. |
Jens Axboe | a68bbdd | 2008-09-24 13:03:33 +0200 | [diff] [blame] | 2349 | */ |
Jens Axboe | f7d7b7a | 2008-09-25 11:37:50 +0200 | [diff] [blame] | 2350 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
Jens Axboe | a68bbdd | 2008-09-24 13:03:33 +0200 | [diff] [blame] | 2351 | return; |
| 2352 | |
Jens Axboe | dd67d05 | 2006-06-21 09:36:18 +0200 | [diff] [blame] | 2353 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2354 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2355 | |
| 2356 | /* |
| 2357 | * idle is disabled, either manually or by past process history |
| 2358 | */ |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2359 | if (!cfq_should_idle(cfqd, cfqq)) { |
| 2360 | /* no queue idling. Check for group idling */ |
| 2361 | if (cfqd->cfq_group_idle) |
| 2362 | group_idle = cfqd->cfq_group_idle; |
| 2363 | else |
| 2364 | return; |
| 2365 | } |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2366 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2367 | /* |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 2368 | * still active requests from this queue, don't idle |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2369 | */ |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 2370 | if (cfqq->dispatched) |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2371 | return; |
| 2372 | |
| 2373 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2374 | * task has exited, don't wait |
| 2375 | */ |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 2376 | cic = cfqd->active_cic; |
Tejun Heo | f6e8d01 | 2012-03-05 13:15:26 -0800 | [diff] [blame] | 2377 | if (!cic || !atomic_read(&cic->icq.ioc->active_ref)) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2378 | return; |
| 2379 | |
Corrado Zoccolo | 355b659 | 2009-10-08 08:43:32 +0200 | [diff] [blame] | 2380 | /* |
| 2381 | * If our average think time is larger than the remaining time |
| 2382 | * slice, then don't idle. This avoids overrunning the allotted |
| 2383 | * time slice. |
| 2384 | */ |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 2385 | if (sample_valid(cic->ttime.ttime_samples) && |
| 2386 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { |
Joe Perches | fd16d26 | 2011-06-13 10:42:49 +0200 | [diff] [blame] | 2387 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 2388 | cic->ttime.ttime_mean); |
Corrado Zoccolo | 355b659 | 2009-10-08 08:43:32 +0200 | [diff] [blame] | 2389 | return; |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 2390 | } |
Corrado Zoccolo | 355b659 | 2009-10-08 08:43:32 +0200 | [diff] [blame] | 2391 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2392 | /* There are other queues in the group, don't do group idle */ |
| 2393 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) |
| 2394 | return; |
| 2395 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2396 | cfq_mark_cfqq_wait_request(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2397 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2398 | if (group_idle) |
| 2399 | sl = cfqd->cfq_group_idle; |
| 2400 | else |
| 2401 | sl = cfqd->cfq_slice_idle; |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 2402 | |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2403 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 2404 | cfq_blkiocg_update_set_idle_time_stats(cfqg_to_blkg(cfqq->cfqg), |
| 2405 | &blkio_policy_cfq); |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2406 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
| 2407 | group_idle ? 1 : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2408 | } |
| 2409 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2410 | /* |
| 2411 | * Move request from internal lists to the request queue dispatch list. |
| 2412 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 2413 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2414 | { |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2415 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 2416 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2417 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2418 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
| 2419 | |
Jeff Moyer | 06d2188 | 2009-09-11 17:08:59 +0200 | [diff] [blame] | 2420 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 2421 | cfq_remove_request(rq); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2422 | cfqq->dispatched++; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2423 | (RQ_CFQG(rq))->dispatched++; |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 2424 | elv_dispatch_sort(q, rq); |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2425 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 2426 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
Vivek Goyal | c4e7893 | 2010-08-23 12:25:03 +0200 | [diff] [blame] | 2427 | cfqq->nr_sectors += blk_rq_sectors(rq); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 2428 | cfq_blkiocg_update_dispatch_stats(cfqg_to_blkg(cfqq->cfqg), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 2429 | &blkio_policy_cfq, blk_rq_bytes(rq), |
| 2430 | rq_data_dir(rq), rq_is_sync(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2431 | } |
| 2432 | |
| 2433 | /* |
| 2434 | * return expired entry, or NULL to just start from scratch in rbtree |
| 2435 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 2436 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2437 | { |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 2438 | struct request *rq = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2439 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2440 | if (cfq_cfqq_fifo_expire(cfqq)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2441 | return NULL; |
Jens Axboe | cb88741 | 2007-01-19 12:01:16 +1100 | [diff] [blame] | 2442 | |
| 2443 | cfq_mark_cfqq_fifo_expire(cfqq); |
| 2444 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2445 | if (list_empty(&cfqq->fifo)) |
| 2446 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2447 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2448 | rq = rq_entry_fifo(cfqq->fifo.next); |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 2449 | if (time_before(jiffies, rq_fifo_time(rq))) |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2450 | rq = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2451 | |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 2452 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2453 | return rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2454 | } |
| 2455 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2456 | static inline int |
| 2457 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2458 | { |
| 2459 | const int base_rq = cfqd->cfq_slice_async_rq; |
| 2460 | |
| 2461 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
| 2462 | |
Namhyung Kim | b9f8ce0 | 2011-05-24 10:23:21 +0200 | [diff] [blame] | 2463 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2464 | } |
| 2465 | |
| 2466 | /* |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2467 | * Must be called with the queue_lock held. |
| 2468 | */ |
| 2469 | static int cfqq_process_refs(struct cfq_queue *cfqq) |
| 2470 | { |
| 2471 | int process_refs, io_refs; |
| 2472 | |
| 2473 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2474 | process_refs = cfqq->ref - io_refs; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2475 | BUG_ON(process_refs < 0); |
| 2476 | return process_refs; |
| 2477 | } |
| 2478 | |
| 2479 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) |
| 2480 | { |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2481 | int process_refs, new_process_refs; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2482 | struct cfq_queue *__cfqq; |
| 2483 | |
Jeff Moyer | c10b61f | 2010-06-17 10:19:11 -0400 | [diff] [blame] | 2484 | /* |
| 2485 | * If there are no process references on the new_cfqq, then it is |
| 2486 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the |
| 2487 | * chain may have dropped their last reference (not just their |
| 2488 | * last process reference). |
| 2489 | */ |
| 2490 | if (!cfqq_process_refs(new_cfqq)) |
| 2491 | return; |
| 2492 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2493 | /* Avoid a circular list and skip interim queue merges */ |
| 2494 | while ((__cfqq = new_cfqq->new_cfqq)) { |
| 2495 | if (__cfqq == cfqq) |
| 2496 | return; |
| 2497 | new_cfqq = __cfqq; |
| 2498 | } |
| 2499 | |
| 2500 | process_refs = cfqq_process_refs(cfqq); |
Jeff Moyer | c10b61f | 2010-06-17 10:19:11 -0400 | [diff] [blame] | 2501 | new_process_refs = cfqq_process_refs(new_cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2502 | /* |
| 2503 | * If the process for the cfqq has gone away, there is no |
| 2504 | * sense in merging the queues. |
| 2505 | */ |
Jeff Moyer | c10b61f | 2010-06-17 10:19:11 -0400 | [diff] [blame] | 2506 | if (process_refs == 0 || new_process_refs == 0) |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2507 | return; |
| 2508 | |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2509 | /* |
| 2510 | * Merge in the direction of the lesser amount of work. |
| 2511 | */ |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2512 | if (new_process_refs >= process_refs) { |
| 2513 | cfqq->new_cfqq = new_cfqq; |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2514 | new_cfqq->ref += process_refs; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2515 | } else { |
| 2516 | new_cfqq->new_cfqq = cfqq; |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2517 | cfqq->ref += new_process_refs; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2518 | } |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2519 | } |
| 2520 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2521 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2522 | struct cfq_group *cfqg, enum wl_prio_t prio) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2523 | { |
| 2524 | struct cfq_queue *queue; |
| 2525 | int i; |
| 2526 | bool key_valid = false; |
| 2527 | unsigned long lowest_key = 0; |
| 2528 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; |
| 2529 | |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2530 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
| 2531 | /* select the one with lowest rb_key */ |
| 2532 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2533 | if (queue && |
| 2534 | (!key_valid || time_before(queue->rb_key, lowest_key))) { |
| 2535 | lowest_key = queue->rb_key; |
| 2536 | cur_best = i; |
| 2537 | key_valid = true; |
| 2538 | } |
| 2539 | } |
| 2540 | |
| 2541 | return cur_best; |
| 2542 | } |
| 2543 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2544 | static void choose_service_tree(struct cfq_data *cfqd, struct cfq_group *cfqg) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2545 | { |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2546 | unsigned slice; |
| 2547 | unsigned count; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2548 | struct cfq_rb_root *st; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2549 | unsigned group_slice; |
Shaohua Li writes | e4ea0c1 | 2010-12-13 14:32:22 +0100 | [diff] [blame] | 2550 | enum wl_prio_t original_prio = cfqd->serving_prio; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2551 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2552 | /* Choose next priority. RT > BE > IDLE */ |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2553 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2554 | cfqd->serving_prio = RT_WORKLOAD; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2555 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2556 | cfqd->serving_prio = BE_WORKLOAD; |
| 2557 | else { |
| 2558 | cfqd->serving_prio = IDLE_WORKLOAD; |
| 2559 | cfqd->workload_expires = jiffies + 1; |
| 2560 | return; |
| 2561 | } |
| 2562 | |
Shaohua Li writes | e4ea0c1 | 2010-12-13 14:32:22 +0100 | [diff] [blame] | 2563 | if (original_prio != cfqd->serving_prio) |
| 2564 | goto new_workload; |
| 2565 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2566 | /* |
| 2567 | * For RT and BE, we have to choose also the type |
| 2568 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload |
| 2569 | * expiration time |
| 2570 | */ |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2571 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2572 | count = st->count; |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2573 | |
| 2574 | /* |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2575 | * check workload expiration, and that we still have other queues ready |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2576 | */ |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2577 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2578 | return; |
| 2579 | |
Shaohua Li writes | e4ea0c1 | 2010-12-13 14:32:22 +0100 | [diff] [blame] | 2580 | new_workload: |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2581 | /* otherwise select new workload type */ |
| 2582 | cfqd->serving_type = |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2583 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
| 2584 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2585 | count = st->count; |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2586 | |
| 2587 | /* |
| 2588 | * the workload slice is computed as a fraction of target latency |
| 2589 | * proportional to the number of queues in that workload, over |
| 2590 | * all the queues in the same priority class |
| 2591 | */ |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2592 | group_slice = cfq_group_slice(cfqd, cfqg); |
| 2593 | |
| 2594 | slice = group_slice * count / |
| 2595 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], |
| 2596 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2597 | |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 2598 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
| 2599 | unsigned int tmp; |
| 2600 | |
| 2601 | /* |
| 2602 | * Async queues are currently system wide. Just taking |
| 2603 | * proportion of queues with-in same group will lead to higher |
| 2604 | * async ratio system wide as generally root group is going |
| 2605 | * to have higher weight. A more accurate thing would be to |
| 2606 | * calculate system wide asnc/sync ratio. |
| 2607 | */ |
| 2608 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); |
| 2609 | tmp = tmp/cfqd->busy_queues; |
| 2610 | slice = min_t(unsigned, slice, tmp); |
| 2611 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2612 | /* async workload slice is scaled down according to |
| 2613 | * the sync/async slice ratio. */ |
| 2614 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 2615 | } else |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2616 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
| 2617 | slice = max(slice, 2 * cfqd->cfq_slice_idle); |
| 2618 | |
| 2619 | slice = max_t(unsigned, slice, CFQ_MIN_TT); |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 2620 | cfq_log(cfqd, "workload slice:%d", slice); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2621 | cfqd->workload_expires = jiffies + slice; |
| 2622 | } |
| 2623 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2624 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
| 2625 | { |
| 2626 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 2627 | struct cfq_group *cfqg; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2628 | |
| 2629 | if (RB_EMPTY_ROOT(&st->rb)) |
| 2630 | return NULL; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 2631 | cfqg = cfq_rb_first_group(st); |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 2632 | update_min_vdisktime(st); |
| 2633 | return cfqg; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2634 | } |
| 2635 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2636 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
| 2637 | { |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2638 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
| 2639 | |
| 2640 | cfqd->serving_group = cfqg; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 2641 | |
| 2642 | /* Restore the workload type data */ |
| 2643 | if (cfqg->saved_workload_slice) { |
| 2644 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; |
| 2645 | cfqd->serving_type = cfqg->saved_workload; |
| 2646 | cfqd->serving_prio = cfqg->saved_serving_prio; |
Gui Jianfeng | 66ae291 | 2009-12-15 10:08:45 +0100 | [diff] [blame] | 2647 | } else |
| 2648 | cfqd->workload_expires = jiffies - 1; |
| 2649 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2650 | choose_service_tree(cfqd, cfqg); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2651 | } |
| 2652 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2653 | /* |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2654 | * Select a queue for service. If we have a current active queue, |
| 2655 | * check whether to continue servicing it, or retrieve and set a new one. |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2656 | */ |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2657 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2658 | { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2659 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2660 | |
| 2661 | cfqq = cfqd->active_queue; |
| 2662 | if (!cfqq) |
| 2663 | goto new_queue; |
| 2664 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2665 | if (!cfqd->rq_queued) |
| 2666 | return NULL; |
Vivek Goyal | c244bb5 | 2009-12-08 17:52:57 -0500 | [diff] [blame] | 2667 | |
| 2668 | /* |
| 2669 | * We were waiting for group to get backlogged. Expire the queue |
| 2670 | */ |
| 2671 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 2672 | goto expire; |
| 2673 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2674 | /* |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2675 | * The active queue has run out of time, expire it and select new. |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2676 | */ |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 2677 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
| 2678 | /* |
| 2679 | * If slice had not expired at the completion of last request |
| 2680 | * we might not have turned on wait_busy flag. Don't expire |
| 2681 | * the queue yet. Allow the group to get backlogged. |
| 2682 | * |
| 2683 | * The very fact that we have used the slice, that means we |
| 2684 | * have been idling all along on this queue and it should be |
| 2685 | * ok to wait for this request to complete. |
| 2686 | */ |
Vivek Goyal | 82bbbf2 | 2009-12-10 19:25:41 +0100 | [diff] [blame] | 2687 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
| 2688 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
| 2689 | cfqq = NULL; |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 2690 | goto keep_queue; |
Vivek Goyal | 82bbbf2 | 2009-12-10 19:25:41 +0100 | [diff] [blame] | 2691 | } else |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2692 | goto check_group_idle; |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 2693 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2694 | |
| 2695 | /* |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2696 | * The active queue has requests and isn't expired, allow it to |
| 2697 | * dispatch. |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2698 | */ |
Jens Axboe | dd67d05 | 2006-06-21 09:36:18 +0200 | [diff] [blame] | 2699 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2700 | goto keep_queue; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2701 | |
| 2702 | /* |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2703 | * If another queue has a request waiting within our mean seek |
| 2704 | * distance, let it run. The expire code will check for close |
| 2705 | * cooperators and put the close queue at the front of the service |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2706 | * tree. If possible, merge the expiring queue with the new cfqq. |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2707 | */ |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 2708 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2709 | if (new_cfqq) { |
| 2710 | if (!cfqq->new_cfqq) |
| 2711 | cfq_setup_merge(cfqq, new_cfqq); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2712 | goto expire; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2713 | } |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2714 | |
| 2715 | /* |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2716 | * No requests pending. If the active queue still has requests in |
| 2717 | * flight or is idling for a new request, allow either of these |
| 2718 | * conditions to happen (or time out) before selecting a new queue. |
| 2719 | */ |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2720 | if (timer_pending(&cfqd->idle_slice_timer)) { |
| 2721 | cfqq = NULL; |
| 2722 | goto keep_queue; |
| 2723 | } |
| 2724 | |
Shaohua Li | 8e1ac66 | 2010-11-08 15:01:04 +0100 | [diff] [blame] | 2725 | /* |
| 2726 | * This is a deep seek queue, but the device is much faster than |
| 2727 | * the queue can deliver, don't idle |
| 2728 | **/ |
| 2729 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && |
| 2730 | (cfq_cfqq_slice_new(cfqq) || |
| 2731 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { |
| 2732 | cfq_clear_cfqq_deep(cfqq); |
| 2733 | cfq_clear_cfqq_idle_window(cfqq); |
| 2734 | } |
| 2735 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2736 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
| 2737 | cfqq = NULL; |
| 2738 | goto keep_queue; |
| 2739 | } |
| 2740 | |
| 2741 | /* |
| 2742 | * If group idle is enabled and there are requests dispatched from |
| 2743 | * this group, wait for requests to complete. |
| 2744 | */ |
| 2745 | check_group_idle: |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 2746 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
| 2747 | cfqq->cfqg->dispatched && |
| 2748 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2749 | cfqq = NULL; |
| 2750 | goto keep_queue; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2751 | } |
| 2752 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2753 | expire: |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2754 | cfq_slice_expired(cfqd, 0); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2755 | new_queue: |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2756 | /* |
| 2757 | * Current queue expired. Check if we have to switch to a new |
| 2758 | * service tree |
| 2759 | */ |
| 2760 | if (!new_cfqq) |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2761 | cfq_choose_cfqg(cfqd); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2762 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2763 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2764 | keep_queue: |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2765 | return cfqq; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2766 | } |
| 2767 | |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 2768 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2769 | { |
| 2770 | int dispatched = 0; |
| 2771 | |
| 2772 | while (cfqq->next_rq) { |
| 2773 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); |
| 2774 | dispatched++; |
| 2775 | } |
| 2776 | |
| 2777 | BUG_ON(!list_empty(&cfqq->fifo)); |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2778 | |
| 2779 | /* By default cfqq is not expired if it is empty. Do it explicitly */ |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2780 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2781 | return dispatched; |
| 2782 | } |
| 2783 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2784 | /* |
| 2785 | * Drain our current requests. Used for barriers and when switching |
| 2786 | * io schedulers on-the-fly. |
| 2787 | */ |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2788 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2789 | { |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 2790 | struct cfq_queue *cfqq; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2791 | int dispatched = 0; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2792 | |
Divyesh Shah | 3440c49 | 2010-04-09 09:29:57 +0200 | [diff] [blame] | 2793 | /* Expire the timeslice of the current active queue first */ |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2794 | cfq_slice_expired(cfqd, 0); |
Divyesh Shah | 3440c49 | 2010-04-09 09:29:57 +0200 | [diff] [blame] | 2795 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
| 2796 | __cfq_set_active_queue(cfqd, cfqq); |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2797 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
Divyesh Shah | 3440c49 | 2010-04-09 09:29:57 +0200 | [diff] [blame] | 2798 | } |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2799 | |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2800 | BUG_ON(cfqd->busy_queues); |
| 2801 | |
Jeff Moyer | 6923715 | 2009-06-12 15:29:30 +0200 | [diff] [blame] | 2802 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2803 | return dispatched; |
| 2804 | } |
| 2805 | |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2806 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
| 2807 | struct cfq_queue *cfqq) |
| 2808 | { |
| 2809 | /* the queue hasn't finished any request, can't estimate */ |
| 2810 | if (cfq_cfqq_slice_new(cfqq)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2811 | return true; |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2812 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
| 2813 | cfqq->slice_end)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2814 | return true; |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2815 | |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2816 | return false; |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2817 | } |
| 2818 | |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2819 | static bool cfq_may_dispatch(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2820 | { |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2821 | unsigned int max_dispatch; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2822 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2823 | /* |
Jens Axboe | 5ad531d | 2009-07-03 12:57:48 +0200 | [diff] [blame] | 2824 | * Drain async requests before we start sync IO |
| 2825 | */ |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 2826 | if (cfq_should_idle(cfqd, cfqq) && cfqd->rq_in_flight[BLK_RW_ASYNC]) |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2827 | return false; |
Jens Axboe | 5ad531d | 2009-07-03 12:57:48 +0200 | [diff] [blame] | 2828 | |
| 2829 | /* |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2830 | * If this is an async queue and we have sync IO in flight, let it wait |
| 2831 | */ |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 2832 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2833 | return false; |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2834 | |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2835 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2836 | if (cfq_class_idle(cfqq)) |
| 2837 | max_dispatch = 1; |
| 2838 | |
| 2839 | /* |
| 2840 | * Does this cfqq already have too much IO in flight? |
| 2841 | */ |
| 2842 | if (cfqq->dispatched >= max_dispatch) { |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2843 | bool promote_sync = false; |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2844 | /* |
| 2845 | * idle queue must always only have a single IO in flight |
| 2846 | */ |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2847 | if (cfq_class_idle(cfqq)) |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2848 | return false; |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2849 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2850 | /* |
Li, Shaohua | c4ade94 | 2011-03-23 08:30:34 +0100 | [diff] [blame] | 2851 | * If there is only one sync queue |
| 2852 | * we can ignore async queue here and give the sync |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2853 | * queue no dispatch limit. The reason is a sync queue can |
| 2854 | * preempt async queue, limiting the sync queue doesn't make |
| 2855 | * sense. This is useful for aiostress test. |
| 2856 | */ |
Li, Shaohua | c4ade94 | 2011-03-23 08:30:34 +0100 | [diff] [blame] | 2857 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
| 2858 | promote_sync = true; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2859 | |
| 2860 | /* |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2861 | * We have other queues, don't allow more IO from this one |
| 2862 | */ |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2863 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
| 2864 | !promote_sync) |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2865 | return false; |
Jens Axboe | 9ede209 | 2007-01-19 12:11:44 +1100 | [diff] [blame] | 2866 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2867 | /* |
Shaohua Li | 474b18c | 2009-12-03 12:58:05 +0100 | [diff] [blame] | 2868 | * Sole queue user, no limit |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 2869 | */ |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2870 | if (cfqd->busy_queues == 1 || promote_sync) |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2871 | max_dispatch = -1; |
| 2872 | else |
| 2873 | /* |
| 2874 | * Normally we start throttling cfqq when cfq_quantum/2 |
| 2875 | * requests have been dispatched. But we can drive |
| 2876 | * deeper queue depths at the beginning of slice |
| 2877 | * subjected to upper limit of cfq_quantum. |
| 2878 | * */ |
| 2879 | max_dispatch = cfqd->cfq_quantum; |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2880 | } |
| 2881 | |
| 2882 | /* |
| 2883 | * Async queues must wait a bit before being allowed dispatch. |
| 2884 | * We also ramp up the dispatch depth gradually for async IO, |
| 2885 | * based on the last sync IO we serviced |
| 2886 | */ |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 2887 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 2888 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2889 | unsigned int depth; |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 2890 | |
Jens Axboe | 61f0c1d | 2009-10-03 19:46:03 +0200 | [diff] [blame] | 2891 | depth = last_sync / cfqd->cfq_slice[1]; |
Jens Axboe | e00c54c | 2009-10-04 20:36:19 +0200 | [diff] [blame] | 2892 | if (!depth && !cfqq->dispatched) |
| 2893 | depth = 1; |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2894 | if (depth < max_dispatch) |
| 2895 | max_dispatch = depth; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2896 | } |
| 2897 | |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2898 | /* |
| 2899 | * If we're below the current max, allow a dispatch |
| 2900 | */ |
| 2901 | return cfqq->dispatched < max_dispatch; |
| 2902 | } |
| 2903 | |
| 2904 | /* |
| 2905 | * Dispatch a request from cfqq, moving them to the request queue |
| 2906 | * dispatch list. |
| 2907 | */ |
| 2908 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2909 | { |
| 2910 | struct request *rq; |
| 2911 | |
| 2912 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
| 2913 | |
| 2914 | if (!cfq_may_dispatch(cfqd, cfqq)) |
| 2915 | return false; |
| 2916 | |
| 2917 | /* |
| 2918 | * follow expired path, else get first next available |
| 2919 | */ |
| 2920 | rq = cfq_check_fifo(cfqq); |
| 2921 | if (!rq) |
| 2922 | rq = cfqq->next_rq; |
| 2923 | |
| 2924 | /* |
| 2925 | * insert request into driver dispatch list |
| 2926 | */ |
| 2927 | cfq_dispatch_insert(cfqd->queue, rq); |
| 2928 | |
| 2929 | if (!cfqd->active_cic) { |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 2930 | struct cfq_io_cq *cic = RQ_CIC(rq); |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2931 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 2932 | atomic_long_inc(&cic->icq.ioc->refcount); |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2933 | cfqd->active_cic = cic; |
| 2934 | } |
| 2935 | |
| 2936 | return true; |
| 2937 | } |
| 2938 | |
| 2939 | /* |
| 2940 | * Find the cfqq that we need to service and move a request from that to the |
| 2941 | * dispatch list |
| 2942 | */ |
| 2943 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
| 2944 | { |
| 2945 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2946 | struct cfq_queue *cfqq; |
| 2947 | |
| 2948 | if (!cfqd->busy_queues) |
| 2949 | return 0; |
| 2950 | |
| 2951 | if (unlikely(force)) |
| 2952 | return cfq_forced_dispatch(cfqd); |
| 2953 | |
| 2954 | cfqq = cfq_select_queue(cfqd); |
| 2955 | if (!cfqq) |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2956 | return 0; |
| 2957 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2958 | /* |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2959 | * Dispatch a request from this cfqq, if it is allowed |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2960 | */ |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2961 | if (!cfq_dispatch_request(cfqd, cfqq)) |
| 2962 | return 0; |
| 2963 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2964 | cfqq->slice_dispatch++; |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 2965 | cfq_clear_cfqq_must_dispatch(cfqq); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2966 | |
| 2967 | /* |
| 2968 | * expire an async queue immediately if it has used up its slice. idle |
| 2969 | * queue always expire after 1 dispatch round. |
| 2970 | */ |
| 2971 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
| 2972 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
| 2973 | cfq_class_idle(cfqq))) { |
| 2974 | cfqq->slice_end = jiffies + 1; |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2975 | cfq_slice_expired(cfqd, 0); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2976 | } |
| 2977 | |
Shan Wei | b217a90 | 2009-09-01 10:06:42 +0200 | [diff] [blame] | 2978 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2979 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2980 | } |
| 2981 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2982 | /* |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 2983 | * task holds one reference to the queue, dropped when task exits. each rq |
| 2984 | * in-flight on this queue also holds a reference, dropped when rq is freed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2985 | * |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 2986 | * Each cfq queue took a reference on the parent group. Drop it now. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2987 | * queue lock must be held here. |
| 2988 | */ |
| 2989 | static void cfq_put_queue(struct cfq_queue *cfqq) |
| 2990 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2991 | struct cfq_data *cfqd = cfqq->cfqd; |
Justin TerAvest | 0bbfeb8 | 2011-03-01 15:05:08 -0500 | [diff] [blame] | 2992 | struct cfq_group *cfqg; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2993 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2994 | BUG_ON(cfqq->ref <= 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2995 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2996 | cfqq->ref--; |
| 2997 | if (cfqq->ref) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2998 | return; |
| 2999 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3000 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3001 | BUG_ON(rb_first(&cfqq->sort_list)); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3002 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 3003 | cfqg = cfqq->cfqg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3004 | |
Jens Axboe | 28f95cbc | 2007-01-19 12:09:53 +1100 | [diff] [blame] | 3005 | if (unlikely(cfqd->active_queue == cfqq)) { |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3006 | __cfq_slice_expired(cfqd, cfqq, 0); |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3007 | cfq_schedule_dispatch(cfqd); |
Jens Axboe | 28f95cbc | 2007-01-19 12:09:53 +1100 | [diff] [blame] | 3008 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3009 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 3010 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3011 | kmem_cache_free(cfq_pool, cfqq); |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 3012 | cfqg_put(cfqg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3013 | } |
| 3014 | |
Shaohua Li | d02a2c0 | 2010-05-25 10:16:53 +0200 | [diff] [blame] | 3015 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3016 | { |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3017 | struct cfq_queue *__cfqq, *next; |
| 3018 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3019 | /* |
| 3020 | * If this queue was scheduled to merge with another queue, be |
| 3021 | * sure to drop the reference taken on that queue (and others in |
| 3022 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. |
| 3023 | */ |
| 3024 | __cfqq = cfqq->new_cfqq; |
| 3025 | while (__cfqq) { |
| 3026 | if (__cfqq == cfqq) { |
| 3027 | WARN(1, "cfqq->new_cfqq loop detected\n"); |
| 3028 | break; |
| 3029 | } |
| 3030 | next = __cfqq->new_cfqq; |
| 3031 | cfq_put_queue(__cfqq); |
| 3032 | __cfqq = next; |
| 3033 | } |
Shaohua Li | d02a2c0 | 2010-05-25 10:16:53 +0200 | [diff] [blame] | 3034 | } |
| 3035 | |
| 3036 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3037 | { |
| 3038 | if (unlikely(cfqq == cfqd->active_queue)) { |
| 3039 | __cfq_slice_expired(cfqd, cfqq, 0); |
| 3040 | cfq_schedule_dispatch(cfqd); |
| 3041 | } |
| 3042 | |
| 3043 | cfq_put_cooperator(cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3044 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3045 | cfq_put_queue(cfqq); |
| 3046 | } |
| 3047 | |
Tejun Heo | 9b84cac | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 3048 | static void cfq_init_icq(struct io_cq *icq) |
| 3049 | { |
| 3050 | struct cfq_io_cq *cic = icq_to_cic(icq); |
| 3051 | |
| 3052 | cic->ttime.last_end_request = jiffies; |
| 3053 | } |
| 3054 | |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3055 | static void cfq_exit_icq(struct io_cq *icq) |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3056 | { |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3057 | struct cfq_io_cq *cic = icq_to_cic(icq); |
Tejun Heo | 283287a | 2011-12-14 00:33:38 +0100 | [diff] [blame] | 3058 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
Fabio Checconi | 4faa3c8 | 2008-04-10 08:28:01 +0200 | [diff] [blame] | 3059 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 3060 | if (cic->cfqq[BLK_RW_ASYNC]) { |
| 3061 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); |
| 3062 | cic->cfqq[BLK_RW_ASYNC] = NULL; |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3063 | } |
| 3064 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 3065 | if (cic->cfqq[BLK_RW_SYNC]) { |
| 3066 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); |
| 3067 | cic->cfqq[BLK_RW_SYNC] = NULL; |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3068 | } |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3069 | } |
| 3070 | |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3071 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct cfq_io_cq *cic) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3072 | { |
| 3073 | struct task_struct *tsk = current; |
| 3074 | int ioprio_class; |
| 3075 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3076 | if (!cfq_cfqq_prio_changed(cfqq)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3077 | return; |
| 3078 | |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3079 | ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3080 | switch (ioprio_class) { |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3081 | default: |
| 3082 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); |
| 3083 | case IOPRIO_CLASS_NONE: |
| 3084 | /* |
Jens Axboe | 6d63c27 | 2008-05-07 09:51:23 +0200 | [diff] [blame] | 3085 | * no prio set, inherit CPU scheduling settings |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3086 | */ |
| 3087 | cfqq->ioprio = task_nice_ioprio(tsk); |
Jens Axboe | 6d63c27 | 2008-05-07 09:51:23 +0200 | [diff] [blame] | 3088 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3089 | break; |
| 3090 | case IOPRIO_CLASS_RT: |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3091 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3092 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
| 3093 | break; |
| 3094 | case IOPRIO_CLASS_BE: |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3095 | cfqq->ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3096 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
| 3097 | break; |
| 3098 | case IOPRIO_CLASS_IDLE: |
| 3099 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; |
| 3100 | cfqq->ioprio = 7; |
| 3101 | cfq_clear_cfqq_idle_window(cfqq); |
| 3102 | break; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3103 | } |
| 3104 | |
| 3105 | /* |
| 3106 | * keep track of original prio settings in case we have to temporarily |
| 3107 | * elevate the priority of this queue |
| 3108 | */ |
| 3109 | cfqq->org_ioprio = cfqq->ioprio; |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3110 | cfq_clear_cfqq_prio_changed(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3111 | } |
| 3112 | |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3113 | static void check_ioprio_changed(struct cfq_io_cq *cic, struct bio *bio) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3114 | { |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3115 | int ioprio = cic->icq.ioc->ioprio; |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 3116 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
Al Viro | 478a82b | 2006-03-18 13:25:24 -0500 | [diff] [blame] | 3117 | struct cfq_queue *cfqq; |
Jens Axboe | 35e6077 | 2006-06-14 09:10:45 +0200 | [diff] [blame] | 3118 | |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3119 | /* |
| 3120 | * Check whether ioprio has changed. The condition may trigger |
| 3121 | * spuriously on a newly created cic but there's no harm. |
| 3122 | */ |
| 3123 | if (unlikely(!cfqd) || likely(cic->ioprio == ioprio)) |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3124 | return; |
| 3125 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 3126 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3127 | if (cfqq) { |
| 3128 | struct cfq_queue *new_cfqq; |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3129 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic, bio, |
| 3130 | GFP_ATOMIC); |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3131 | if (new_cfqq) { |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 3132 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3133 | cfq_put_queue(cfqq); |
| 3134 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3135 | } |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3136 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 3137 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3138 | if (cfqq) |
| 3139 | cfq_mark_cfqq_prio_changed(cfqq); |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3140 | |
| 3141 | cic->ioprio = ioprio; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3142 | } |
| 3143 | |
Jens Axboe | d5036d7 | 2009-06-26 10:44:34 +0200 | [diff] [blame] | 3144 | static void cfq_init_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3145 | pid_t pid, bool is_sync) |
Jens Axboe | d5036d7 | 2009-06-26 10:44:34 +0200 | [diff] [blame] | 3146 | { |
| 3147 | RB_CLEAR_NODE(&cfqq->rb_node); |
| 3148 | RB_CLEAR_NODE(&cfqq->p_node); |
| 3149 | INIT_LIST_HEAD(&cfqq->fifo); |
| 3150 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 3151 | cfqq->ref = 0; |
Jens Axboe | d5036d7 | 2009-06-26 10:44:34 +0200 | [diff] [blame] | 3152 | cfqq->cfqd = cfqd; |
| 3153 | |
| 3154 | cfq_mark_cfqq_prio_changed(cfqq); |
| 3155 | |
| 3156 | if (is_sync) { |
| 3157 | if (!cfq_class_idle(cfqq)) |
| 3158 | cfq_mark_cfqq_idle_window(cfqq); |
| 3159 | cfq_mark_cfqq_sync(cfqq); |
| 3160 | } |
| 3161 | cfqq->pid = pid; |
| 3162 | } |
| 3163 | |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3164 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3165 | static void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3166 | { |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 3167 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3168 | struct cfq_queue *sync_cfqq; |
| 3169 | uint64_t id; |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3170 | |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3171 | rcu_read_lock(); |
| 3172 | id = bio_blkio_cgroup(bio)->id; |
| 3173 | rcu_read_unlock(); |
| 3174 | |
| 3175 | /* |
| 3176 | * Check whether blkcg has changed. The condition may trigger |
| 3177 | * spuriously on a newly created cic but there's no harm. |
| 3178 | */ |
| 3179 | if (unlikely(!cfqd) || likely(cic->blkcg_id == id)) |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3180 | return; |
| 3181 | |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3182 | sync_cfqq = cic_to_cfqq(cic, 1); |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3183 | if (sync_cfqq) { |
| 3184 | /* |
| 3185 | * Drop reference to sync queue. A new sync queue will be |
| 3186 | * assigned in new group upon arrival of a fresh request. |
| 3187 | */ |
| 3188 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); |
| 3189 | cic_set_cfqq(cic, NULL, 1); |
| 3190 | cfq_put_queue(sync_cfqq); |
| 3191 | } |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3192 | |
| 3193 | cic->blkcg_id = id; |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3194 | } |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3195 | #else |
| 3196 | static inline void check_blkcg_changed(struct cfq_io_cq *cic, struct bio *bio) { } |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3197 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 3198 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3199 | static struct cfq_queue * |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3200 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
| 3201 | struct bio *bio, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3202 | { |
Tejun Heo | 0a5a7d0 | 2012-03-05 13:15:02 -0800 | [diff] [blame] | 3203 | struct blkio_cgroup *blkcg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3204 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 3205 | struct cfq_group *cfqg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3206 | |
| 3207 | retry: |
Tejun Heo | 2a7f124 | 2012-03-05 13:15:01 -0800 | [diff] [blame] | 3208 | rcu_read_lock(); |
| 3209 | |
Tejun Heo | 4f85cb9 | 2012-03-05 13:15:28 -0800 | [diff] [blame] | 3210 | blkcg = bio_blkio_cgroup(bio); |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 3211 | cfqg = cfq_lookup_create_cfqg(cfqd, blkcg); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3212 | cfqq = cic_to_cfqq(cic, is_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3213 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3214 | /* |
| 3215 | * Always try a new alloc if we fell back to the OOM cfqq |
| 3216 | * originally, since it should just be a temporary situation. |
| 3217 | */ |
| 3218 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
| 3219 | cfqq = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3220 | if (new_cfqq) { |
| 3221 | cfqq = new_cfqq; |
| 3222 | new_cfqq = NULL; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3223 | } else if (gfp_mask & __GFP_WAIT) { |
Tejun Heo | 2a7f124 | 2012-03-05 13:15:01 -0800 | [diff] [blame] | 3224 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3225 | spin_unlock_irq(cfqd->queue->queue_lock); |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 3226 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3227 | gfp_mask | __GFP_ZERO, |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 3228 | cfqd->queue->node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3229 | spin_lock_irq(cfqd->queue->queue_lock); |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3230 | if (new_cfqq) |
| 3231 | goto retry; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3232 | } else { |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 3233 | cfqq = kmem_cache_alloc_node(cfq_pool, |
| 3234 | gfp_mask | __GFP_ZERO, |
| 3235 | cfqd->queue->node); |
Kiyoshi Ueda | db3b584 | 2005-06-17 16:15:10 +0200 | [diff] [blame] | 3236 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3237 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3238 | if (cfqq) { |
| 3239 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3240 | cfq_init_prio_data(cfqq, cic); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 3241 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3242 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
| 3243 | } else |
| 3244 | cfqq = &cfqd->oom_cfqq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3245 | } |
| 3246 | |
| 3247 | if (new_cfqq) |
| 3248 | kmem_cache_free(cfq_pool, new_cfqq); |
| 3249 | |
Tejun Heo | 2a7f124 | 2012-03-05 13:15:01 -0800 | [diff] [blame] | 3250 | rcu_read_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3251 | return cfqq; |
| 3252 | } |
| 3253 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3254 | static struct cfq_queue ** |
| 3255 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) |
| 3256 | { |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3257 | switch (ioprio_class) { |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3258 | case IOPRIO_CLASS_RT: |
| 3259 | return &cfqd->async_cfqq[0][ioprio]; |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3260 | case IOPRIO_CLASS_NONE: |
| 3261 | ioprio = IOPRIO_NORM; |
| 3262 | /* fall through */ |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3263 | case IOPRIO_CLASS_BE: |
| 3264 | return &cfqd->async_cfqq[1][ioprio]; |
| 3265 | case IOPRIO_CLASS_IDLE: |
| 3266 | return &cfqd->async_idle_cfqq; |
| 3267 | default: |
| 3268 | BUG(); |
| 3269 | } |
| 3270 | } |
| 3271 | |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3272 | static struct cfq_queue * |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3273 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct cfq_io_cq *cic, |
Tejun Heo | 4f85cb9 | 2012-03-05 13:15:28 -0800 | [diff] [blame] | 3274 | struct bio *bio, gfp_t gfp_mask) |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3275 | { |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3276 | const int ioprio_class = IOPRIO_PRIO_CLASS(cic->ioprio); |
| 3277 | const int ioprio = IOPRIO_PRIO_DATA(cic->ioprio); |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3278 | struct cfq_queue **async_cfqq = NULL; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3279 | struct cfq_queue *cfqq = NULL; |
| 3280 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3281 | if (!is_sync) { |
| 3282 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); |
| 3283 | cfqq = *async_cfqq; |
| 3284 | } |
| 3285 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3286 | if (!cfqq) |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3287 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, cic, bio, gfp_mask); |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3288 | |
| 3289 | /* |
| 3290 | * pin the queue now that it's allocated, scheduler exit will prune it |
| 3291 | */ |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3292 | if (!is_sync && !(*async_cfqq)) { |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 3293 | cfqq->ref++; |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3294 | *async_cfqq = cfqq; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3295 | } |
| 3296 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 3297 | cfqq->ref++; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3298 | return cfqq; |
| 3299 | } |
| 3300 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3301 | static void |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3302 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3303 | { |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3304 | unsigned long elapsed = jiffies - ttime->last_end_request; |
| 3305 | elapsed = min(elapsed, 2UL * slice_idle); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3306 | |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3307 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
| 3308 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; |
| 3309 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; |
| 3310 | } |
| 3311 | |
| 3312 | static void |
| 3313 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3314 | struct cfq_io_cq *cic) |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3315 | { |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3316 | if (cfq_cfqq_sync(cfqq)) { |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3317 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3318 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
| 3319 | cfqd->cfq_slice_idle); |
| 3320 | } |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 3321 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3322 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); |
| 3323 | #endif |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3324 | } |
| 3325 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3326 | static void |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 3327 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3328 | struct request *rq) |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3329 | { |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 3330 | sector_t sdist = 0; |
Corrado Zoccolo | 41647e7 | 2010-02-27 19:45:40 +0100 | [diff] [blame] | 3331 | sector_t n_sec = blk_rq_sectors(rq); |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 3332 | if (cfqq->last_request_pos) { |
| 3333 | if (cfqq->last_request_pos < blk_rq_pos(rq)) |
| 3334 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; |
| 3335 | else |
| 3336 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); |
| 3337 | } |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3338 | |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 3339 | cfqq->seek_history <<= 1; |
Corrado Zoccolo | 41647e7 | 2010-02-27 19:45:40 +0100 | [diff] [blame] | 3340 | if (blk_queue_nonrot(cfqd->queue)) |
| 3341 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); |
| 3342 | else |
| 3343 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3344 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3345 | |
| 3346 | /* |
| 3347 | * Disable idle window if the process thinks too long or seeks so much that |
| 3348 | * it doesn't matter |
| 3349 | */ |
| 3350 | static void |
| 3351 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3352 | struct cfq_io_cq *cic) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3353 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3354 | int old_idle, enable_idle; |
Jens Axboe | 1be92f2 | 2007-04-19 14:32:26 +0200 | [diff] [blame] | 3355 | |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 3356 | /* |
| 3357 | * Don't idle for async or idle io prio class |
| 3358 | */ |
| 3359 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) |
Jens Axboe | 1be92f2 | 2007-04-19 14:32:26 +0200 | [diff] [blame] | 3360 | return; |
| 3361 | |
Jens Axboe | c265a7f | 2008-06-26 13:49:33 +0200 | [diff] [blame] | 3362 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3363 | |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3364 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
| 3365 | cfq_mark_cfqq_deep(cfqq); |
| 3366 | |
Corrado Zoccolo | 749ef9f | 2010-09-20 15:24:50 +0200 | [diff] [blame] | 3367 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
| 3368 | enable_idle = 0; |
Tejun Heo | f6e8d01 | 2012-03-05 13:15:26 -0800 | [diff] [blame] | 3369 | else if (!atomic_read(&cic->icq.ioc->active_ref) || |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3370 | !cfqd->cfq_slice_idle || |
| 3371 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3372 | enable_idle = 0; |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3373 | else if (sample_valid(cic->ttime.ttime_samples)) { |
| 3374 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3375 | enable_idle = 0; |
| 3376 | else |
| 3377 | enable_idle = 1; |
| 3378 | } |
| 3379 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3380 | if (old_idle != enable_idle) { |
| 3381 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); |
| 3382 | if (enable_idle) |
| 3383 | cfq_mark_cfqq_idle_window(cfqq); |
| 3384 | else |
| 3385 | cfq_clear_cfqq_idle_window(cfqq); |
| 3386 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3387 | } |
| 3388 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3389 | /* |
| 3390 | * Check if new_cfqq should preempt the currently active queue. Return 0 for |
| 3391 | * no or if we aren't sure, a 1 will cause a preempt. |
| 3392 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3393 | static bool |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3394 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3395 | struct request *rq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3396 | { |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3397 | struct cfq_queue *cfqq; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3398 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3399 | cfqq = cfqd->active_queue; |
| 3400 | if (!cfqq) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3401 | return false; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3402 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3403 | if (cfq_class_idle(new_cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3404 | return false; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3405 | |
| 3406 | if (cfq_class_idle(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3407 | return true; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3408 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3409 | /* |
Divyesh Shah | 875feb6 | 2010-01-06 18:58:20 -0800 | [diff] [blame] | 3410 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. |
| 3411 | */ |
| 3412 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) |
| 3413 | return false; |
| 3414 | |
| 3415 | /* |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 3416 | * if the new request is sync, but the currently running queue is |
| 3417 | * not, let the sync request have priority. |
| 3418 | */ |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3419 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3420 | return true; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3421 | |
Vivek Goyal | 8682e1f | 2009-12-03 12:59:50 -0500 | [diff] [blame] | 3422 | if (new_cfqq->cfqg != cfqq->cfqg) |
| 3423 | return false; |
| 3424 | |
| 3425 | if (cfq_slice_used(cfqq)) |
| 3426 | return true; |
| 3427 | |
| 3428 | /* Allow preemption only if we are idling on sync-noidle tree */ |
| 3429 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && |
| 3430 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
| 3431 | new_cfqq->service_tree->count == 2 && |
| 3432 | RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 3433 | return true; |
| 3434 | |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 3435 | /* |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 3436 | * So both queues are sync. Let the new request get disk time if |
| 3437 | * it's a metadata request and the current queue is doing regular IO. |
| 3438 | */ |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 3439 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 3440 | return true; |
| 3441 | |
| 3442 | /* |
Divyesh Shah | 3a9a3f6 | 2009-01-30 12:46:41 +0100 | [diff] [blame] | 3443 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. |
| 3444 | */ |
| 3445 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3446 | return true; |
Divyesh Shah | 3a9a3f6 | 2009-01-30 12:46:41 +0100 | [diff] [blame] | 3447 | |
Shaohua Li | d2d59e1 | 2010-11-08 15:01:03 +0100 | [diff] [blame] | 3448 | /* An idle queue should not be idle now for some reason */ |
| 3449 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) |
| 3450 | return true; |
| 3451 | |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3452 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3453 | return false; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3454 | |
| 3455 | /* |
| 3456 | * if this request is as-good as one we would expect from the |
| 3457 | * current cfqq, let it preempt |
| 3458 | */ |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 3459 | if (cfq_rq_close(cfqd, cfqq, rq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3460 | return true; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3461 | |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3462 | return false; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3463 | } |
| 3464 | |
| 3465 | /* |
| 3466 | * cfqq preempts the active queue. if we allowed preempt with no slice left, |
| 3467 | * let it have half of its nominal slice. |
| 3468 | */ |
| 3469 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3470 | { |
Shaohua Li | df0793a | 2012-01-19 09:20:09 +0100 | [diff] [blame] | 3471 | enum wl_type_t old_type = cfqq_type(cfqd->active_queue); |
| 3472 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3473 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
Shaohua Li | df0793a | 2012-01-19 09:20:09 +0100 | [diff] [blame] | 3474 | cfq_slice_expired(cfqd, 1); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3475 | |
Jens Axboe | bf57225 | 2006-07-19 20:29:12 +0200 | [diff] [blame] | 3476 | /* |
Shaohua Li | f8ae6e3 | 2011-01-14 08:41:02 +0100 | [diff] [blame] | 3477 | * workload type is changed, don't save slice, otherwise preempt |
| 3478 | * doesn't happen |
| 3479 | */ |
Shaohua Li | df0793a | 2012-01-19 09:20:09 +0100 | [diff] [blame] | 3480 | if (old_type != cfqq_type(cfqq)) |
Shaohua Li | f8ae6e3 | 2011-01-14 08:41:02 +0100 | [diff] [blame] | 3481 | cfqq->cfqg->saved_workload_slice = 0; |
| 3482 | |
| 3483 | /* |
Jens Axboe | bf57225 | 2006-07-19 20:29:12 +0200 | [diff] [blame] | 3484 | * Put the new queue at the front of the of the current list, |
| 3485 | * so we know that it will be selected next. |
| 3486 | */ |
| 3487 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 3488 | |
| 3489 | cfq_service_tree_add(cfqd, cfqq, 1); |
Justin TerAvest | eda5e0c | 2011-03-22 21:26:49 +0100 | [diff] [blame] | 3490 | |
Justin TerAvest | 62a37f6 | 2011-03-23 08:25:44 +0100 | [diff] [blame] | 3491 | cfqq->slice_end = 0; |
| 3492 | cfq_mark_cfqq_slice_new(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3493 | } |
| 3494 | |
| 3495 | /* |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3496 | * Called when a new fs request (rq) is added (to cfqq). Check if there's |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3497 | * something we should do about it |
| 3498 | */ |
| 3499 | static void |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3500 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3501 | struct request *rq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3502 | { |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3503 | struct cfq_io_cq *cic = RQ_CIC(rq); |
Jens Axboe | 12e9fdd | 2006-06-01 10:09:56 +0200 | [diff] [blame] | 3504 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3505 | cfqd->rq_queued++; |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 3506 | if (rq->cmd_flags & REQ_PRIO) |
| 3507 | cfqq->prio_pending++; |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 3508 | |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3509 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 3510 | cfq_update_io_seektime(cfqd, cfqq, rq); |
Jens Axboe | 9c2c38a | 2005-08-24 14:57:54 +0200 | [diff] [blame] | 3511 | cfq_update_idle_window(cfqd, cfqq, cic); |
| 3512 | |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 3513 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3514 | |
| 3515 | if (cfqq == cfqd->active_queue) { |
| 3516 | /* |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 3517 | * Remember that we saw a request from this process, but |
| 3518 | * don't start queuing just yet. Otherwise we risk seeing lots |
| 3519 | * of tiny requests, because we disrupt the normal plugging |
Jens Axboe | d6ceb25 | 2009-04-14 14:18:16 +0200 | [diff] [blame] | 3520 | * and merging. If the request is already larger than a single |
| 3521 | * page, let it rip immediately. For that case we assume that |
Jens Axboe | 2d87072 | 2009-04-15 12:12:46 +0200 | [diff] [blame] | 3522 | * merging is already done. Ditto for a busy system that |
| 3523 | * has other work pending, don't risk delaying until the |
| 3524 | * idle timer unplug to continue working. |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3525 | */ |
Jens Axboe | d6ceb25 | 2009-04-14 14:18:16 +0200 | [diff] [blame] | 3526 | if (cfq_cfqq_wait_request(cfqq)) { |
Jens Axboe | 2d87072 | 2009-04-15 12:12:46 +0200 | [diff] [blame] | 3527 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
| 3528 | cfqd->busy_queues > 1) { |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 3529 | cfq_del_timer(cfqd, cfqq); |
Gui Jianfeng | 554554f | 2009-12-10 09:38:39 +0100 | [diff] [blame] | 3530 | cfq_clear_cfqq_wait_request(cfqq); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 3531 | __blk_run_queue(cfqd->queue); |
Divyesh Shah | a11cdaa | 2010-04-13 19:59:17 +0200 | [diff] [blame] | 3532 | } else { |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 3533 | cfq_blkiocg_update_idle_time_stats( |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 3534 | cfqg_to_blkg(cfqq->cfqg), |
| 3535 | &blkio_policy_cfq); |
Vivek Goyal | bf791937 | 2009-12-03 12:59:37 -0500 | [diff] [blame] | 3536 | cfq_mark_cfqq_must_dispatch(cfqq); |
Divyesh Shah | a11cdaa | 2010-04-13 19:59:17 +0200 | [diff] [blame] | 3537 | } |
Jens Axboe | d6ceb25 | 2009-04-14 14:18:16 +0200 | [diff] [blame] | 3538 | } |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3539 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3540 | /* |
| 3541 | * not the active queue - expire current slice if it is |
| 3542 | * idle and has expired it's mean thinktime or this new queue |
Divyesh Shah | 3a9a3f6 | 2009-01-30 12:46:41 +0100 | [diff] [blame] | 3543 | * has some old slice time left and is of higher priority or |
| 3544 | * this new queue is RT and the current one is BE |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3545 | */ |
| 3546 | cfq_preempt_queue(cfqd, cfqq); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 3547 | __blk_run_queue(cfqd->queue); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3548 | } |
| 3549 | } |
| 3550 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3551 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3552 | { |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3553 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3554 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3555 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3556 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3557 | cfq_init_prio_data(cfqq, RQ_CIC(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3558 | |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 3559 | rq_set_fifo_time(rq, jiffies + cfqd->cfq_fifo_expire[rq_is_sync(rq)]); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3560 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 3561 | cfq_add_rq_rb(rq); |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 3562 | cfq_blkiocg_update_io_add_stats(cfqg_to_blkg(RQ_CFQG(rq)), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 3563 | &blkio_policy_cfq, |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 3564 | cfqg_to_blkg(cfqd->serving_group), |
| 3565 | rq_data_dir(rq), rq_is_sync(rq)); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3566 | cfq_rq_enqueued(cfqd, cfqq, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3567 | } |
| 3568 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3569 | /* |
| 3570 | * Update hw_tag based on peak queue depth over 50 samples under |
| 3571 | * sufficient load. |
| 3572 | */ |
| 3573 | static void cfq_update_hw_tag(struct cfq_data *cfqd) |
| 3574 | { |
Shaohua Li | 1a1238a | 2009-10-27 08:46:23 +0100 | [diff] [blame] | 3575 | struct cfq_queue *cfqq = cfqd->active_queue; |
| 3576 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3577 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
| 3578 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 3579 | |
| 3580 | if (cfqd->hw_tag == 1) |
| 3581 | return; |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3582 | |
| 3583 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3584 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3585 | return; |
| 3586 | |
Shaohua Li | 1a1238a | 2009-10-27 08:46:23 +0100 | [diff] [blame] | 3587 | /* |
| 3588 | * If active queue hasn't enough requests and can idle, cfq might not |
| 3589 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this |
| 3590 | * case |
| 3591 | */ |
| 3592 | if (cfqq && cfq_cfqq_idle_window(cfqq) && |
| 3593 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3594 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
Shaohua Li | 1a1238a | 2009-10-27 08:46:23 +0100 | [diff] [blame] | 3595 | return; |
| 3596 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3597 | if (cfqd->hw_tag_samples++ < 50) |
| 3598 | return; |
| 3599 | |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 3600 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3601 | cfqd->hw_tag = 1; |
| 3602 | else |
| 3603 | cfqd->hw_tag = 0; |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3604 | } |
| 3605 | |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3606 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3607 | { |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3608 | struct cfq_io_cq *cic = cfqd->active_cic; |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3609 | |
Justin TerAvest | 02a8f01 | 2011-02-09 14:20:03 +0100 | [diff] [blame] | 3610 | /* If the queue already has requests, don't wait */ |
| 3611 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 3612 | return false; |
| 3613 | |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3614 | /* If there are other queues in the group, don't wait */ |
| 3615 | if (cfqq->cfqg->nr_cfqq > 1) |
| 3616 | return false; |
| 3617 | |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 3618 | /* the only queue in the group, but think time is big */ |
| 3619 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) |
| 3620 | return false; |
| 3621 | |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3622 | if (cfq_slice_used(cfqq)) |
| 3623 | return true; |
| 3624 | |
| 3625 | /* if slice left is less than think time, wait busy */ |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3626 | if (cic && sample_valid(cic->ttime.ttime_samples) |
| 3627 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3628 | return true; |
| 3629 | |
| 3630 | /* |
| 3631 | * If think times is less than a jiffy than ttime_mean=0 and above |
| 3632 | * will not be true. It might happen that slice has not expired yet |
| 3633 | * but will expire soon (4-5 ns) during select_queue(). To cover the |
| 3634 | * case where think time is less than a jiffy, mark the queue wait |
| 3635 | * busy if only 1 jiffy is left in the slice. |
| 3636 | */ |
| 3637 | if (cfqq->slice_end - jiffies == 1) |
| 3638 | return true; |
| 3639 | |
| 3640 | return false; |
| 3641 | } |
| 3642 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3643 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3644 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3645 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3646 | struct cfq_data *cfqd = cfqq->cfqd; |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 3647 | const int sync = rq_is_sync(rq); |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3648 | unsigned long now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3649 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3650 | now = jiffies; |
Christoph Hellwig | 33659eb | 2010-08-07 18:17:56 +0200 | [diff] [blame] | 3651 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
| 3652 | !!(rq->cmd_flags & REQ_NOIDLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3653 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3654 | cfq_update_hw_tag(cfqd); |
| 3655 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3656 | WARN_ON(!cfqd->rq_in_driver); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3657 | WARN_ON(!cfqq->dispatched); |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3658 | cfqd->rq_in_driver--; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3659 | cfqq->dispatched--; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 3660 | (RQ_CFQG(rq))->dispatched--; |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 3661 | cfq_blkiocg_update_completion_stats(cfqg_to_blkg(cfqq->cfqg), |
Tejun Heo | c176826 | 2012-03-05 13:15:17 -0800 | [diff] [blame] | 3662 | &blkio_policy_cfq, rq_start_time_ns(rq), |
| 3663 | rq_io_start_time_ns(rq), rq_data_dir(rq), |
| 3664 | rq_is_sync(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3665 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3666 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 3667 | |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 3668 | if (sync) { |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3669 | struct cfq_rb_root *service_tree; |
| 3670 | |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3671 | RQ_CIC(rq)->ttime.last_end_request = now; |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3672 | |
| 3673 | if (cfq_cfqq_on_rr(cfqq)) |
| 3674 | service_tree = cfqq->service_tree; |
| 3675 | else |
| 3676 | service_tree = service_tree_for(cfqq->cfqg, |
| 3677 | cfqq_prio(cfqq), cfqq_type(cfqq)); |
| 3678 | service_tree->ttime.last_end_request = now; |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 3679 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
| 3680 | cfqd->last_delayed_sync = now; |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 3681 | } |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3682 | |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 3683 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3684 | cfqq->cfqg->ttime.last_end_request = now; |
| 3685 | #endif |
| 3686 | |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3687 | /* |
| 3688 | * If this is the active queue, check if it needs to be expired, |
| 3689 | * or if we want to idle in case it has no pending requests. |
| 3690 | */ |
| 3691 | if (cfqd->active_queue == cfqq) { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 3692 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
| 3693 | |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 3694 | if (cfq_cfqq_slice_new(cfqq)) { |
| 3695 | cfq_set_prio_slice(cfqd, cfqq); |
| 3696 | cfq_clear_cfqq_slice_new(cfqq); |
| 3697 | } |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3698 | |
| 3699 | /* |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3700 | * Should we wait for next request to come in before we expire |
| 3701 | * the queue. |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3702 | */ |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3703 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 3704 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
| 3705 | if (!cfqd->cfq_slice_idle) |
| 3706 | extend_sl = cfqd->cfq_group_idle; |
| 3707 | cfqq->slice_end = jiffies + extend_sl; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3708 | cfq_mark_cfqq_wait_busy(cfqq); |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 3709 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3710 | } |
| 3711 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 3712 | /* |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3713 | * Idling is not enabled on: |
| 3714 | * - expired queues |
| 3715 | * - idle-priority queues |
| 3716 | * - async queues |
| 3717 | * - queues with still some requests queued |
| 3718 | * - when there is a close cooperator |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 3719 | */ |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 3720 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3721 | cfq_slice_expired(cfqd, 1); |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3722 | else if (sync && cfqq_empty && |
| 3723 | !cfq_close_cooperator(cfqd, cfqq)) { |
Corrado Zoccolo | 749ef9f | 2010-09-20 15:24:50 +0200 | [diff] [blame] | 3724 | cfq_arm_slice_timer(cfqd); |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3725 | } |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3726 | } |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3727 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3728 | if (!cfqd->rq_in_driver) |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3729 | cfq_schedule_dispatch(cfqd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3730 | } |
| 3731 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3732 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3733 | { |
Jens Axboe | 1b379d8 | 2009-08-11 08:26:11 +0200 | [diff] [blame] | 3734 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3735 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3736 | return ELV_MQUEUE_MUST; |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3737 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3738 | |
| 3739 | return ELV_MQUEUE_MAY; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3740 | } |
| 3741 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3742 | static int cfq_may_queue(struct request_queue *q, int rw) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3743 | { |
| 3744 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 3745 | struct task_struct *tsk = current; |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3746 | struct cfq_io_cq *cic; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3747 | struct cfq_queue *cfqq; |
| 3748 | |
| 3749 | /* |
| 3750 | * don't force setup of a queue from here, as a call to may_queue |
| 3751 | * does not necessarily imply that a request actually will be queued. |
| 3752 | * so just lookup a possibly existing queue, or return 'may queue' |
| 3753 | * if that fails |
| 3754 | */ |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3755 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3756 | if (!cic) |
| 3757 | return ELV_MQUEUE_MAY; |
| 3758 | |
Jens Axboe | b0b78f8 | 2009-04-08 10:56:08 +0200 | [diff] [blame] | 3759 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3760 | if (cfqq) { |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3761 | cfq_init_prio_data(cfqq, cic); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3762 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3763 | return __cfq_may_queue(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3764 | } |
| 3765 | |
| 3766 | return ELV_MQUEUE_MAY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3767 | } |
| 3768 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3769 | /* |
| 3770 | * queue lock held here |
| 3771 | */ |
Jens Axboe | bb37b94 | 2006-12-01 10:42:33 +0100 | [diff] [blame] | 3772 | static void cfq_put_request(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3773 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3774 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3775 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3776 | if (cfqq) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3777 | const int rw = rq_data_dir(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3778 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3779 | BUG_ON(!cfqq->allocated[rw]); |
| 3780 | cfqq->allocated[rw]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3781 | |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 3782 | /* Put down rq reference on cfqg */ |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 3783 | cfqg_put(RQ_CFQG(rq)); |
Tejun Heo | a612fdd | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3784 | rq->elv.priv[0] = NULL; |
| 3785 | rq->elv.priv[1] = NULL; |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 3786 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3787 | cfq_put_queue(cfqq); |
| 3788 | } |
| 3789 | } |
| 3790 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3791 | static struct cfq_queue * |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3792 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_cq *cic, |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3793 | struct cfq_queue *cfqq) |
| 3794 | { |
| 3795 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); |
| 3796 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 3797 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3798 | cfq_put_queue(cfqq); |
| 3799 | return cic_to_cfqq(cic, 1); |
| 3800 | } |
| 3801 | |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3802 | /* |
| 3803 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this |
| 3804 | * was the last process referring to said cfqq. |
| 3805 | */ |
| 3806 | static struct cfq_queue * |
Tejun Heo | c586980 | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3807 | split_cfqq(struct cfq_io_cq *cic, struct cfq_queue *cfqq) |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3808 | { |
| 3809 | if (cfqq_process_refs(cfqq) == 1) { |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3810 | cfqq->pid = current->pid; |
| 3811 | cfq_clear_cfqq_coop(cfqq); |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 3812 | cfq_clear_cfqq_split_coop(cfqq); |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3813 | return cfqq; |
| 3814 | } |
| 3815 | |
| 3816 | cic_set_cfqq(cic, NULL, 1); |
Shaohua Li | d02a2c0 | 2010-05-25 10:16:53 +0200 | [diff] [blame] | 3817 | |
| 3818 | cfq_put_cooperator(cfqq); |
| 3819 | |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3820 | cfq_put_queue(cfqq); |
| 3821 | return NULL; |
| 3822 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3823 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3824 | * Allocate cfq data structures associated with this request. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3825 | */ |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3826 | static int |
Tejun Heo | 852c788 | 2012-03-05 13:15:27 -0800 | [diff] [blame] | 3827 | cfq_set_request(struct request_queue *q, struct request *rq, struct bio *bio, |
| 3828 | gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3829 | { |
| 3830 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 3831 | struct cfq_io_cq *cic = icq_to_cic(rq->elv.icq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3832 | const int rw = rq_data_dir(rq); |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3833 | const bool is_sync = rq_is_sync(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3834 | struct cfq_queue *cfqq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3835 | |
| 3836 | might_sleep_if(gfp_mask & __GFP_WAIT); |
| 3837 | |
Tejun Heo | 216284c | 2011-12-14 00:33:38 +0100 | [diff] [blame] | 3838 | spin_lock_irq(q->queue_lock); |
Tejun Heo | f1f8cc9 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 3839 | |
Tejun Heo | 598971b | 2012-03-19 15:10:58 -0700 | [diff] [blame] | 3840 | check_ioprio_changed(cic, bio); |
| 3841 | check_blkcg_changed(cic, bio); |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3842 | new_queue: |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3843 | cfqq = cic_to_cfqq(cic, is_sync); |
Vivek Goyal | 32f2e80 | 2009-07-09 22:13:16 +0200 | [diff] [blame] | 3844 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
Tejun Heo | abede6d | 2012-03-19 15:10:57 -0700 | [diff] [blame] | 3845 | cfqq = cfq_get_queue(cfqd, is_sync, cic, bio, gfp_mask); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3846 | cic_set_cfqq(cic, cfqq, is_sync); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3847 | } else { |
| 3848 | /* |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3849 | * If the queue was seeky for too long, break it apart. |
| 3850 | */ |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 3851 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3852 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
| 3853 | cfqq = split_cfqq(cic, cfqq); |
| 3854 | if (!cfqq) |
| 3855 | goto new_queue; |
| 3856 | } |
| 3857 | |
| 3858 | /* |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3859 | * Check to see if this queue is scheduled to merge with |
| 3860 | * another, closely cooperating queue. The merging of |
| 3861 | * queues happens here as it must be done in process context. |
| 3862 | * The reference on new_cfqq was taken in merge_cfqqs. |
| 3863 | */ |
| 3864 | if (cfqq->new_cfqq) |
| 3865 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3866 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3867 | |
| 3868 | cfqq->allocated[rw]++; |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3869 | |
Jens Axboe | 6fae9c2 | 2011-03-01 15:04:39 -0500 | [diff] [blame] | 3870 | cfqq->ref++; |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 3871 | cfqg_get(cfqq->cfqg); |
Tejun Heo | a612fdd | 2011-12-14 00:33:41 +0100 | [diff] [blame] | 3872 | rq->elv.priv[0] = cfqq; |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 3873 | rq->elv.priv[1] = cfqq->cfqg; |
Tejun Heo | 216284c | 2011-12-14 00:33:38 +0100 | [diff] [blame] | 3874 | spin_unlock_irq(q->queue_lock); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3875 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3876 | } |
| 3877 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3878 | static void cfq_kick_queue(struct work_struct *work) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3879 | { |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3880 | struct cfq_data *cfqd = |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3881 | container_of(work, struct cfq_data, unplug_work); |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3882 | struct request_queue *q = cfqd->queue; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3883 | |
Jens Axboe | 40bb54d | 2009-04-15 12:11:10 +0200 | [diff] [blame] | 3884 | spin_lock_irq(q->queue_lock); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 3885 | __blk_run_queue(cfqd->queue); |
Jens Axboe | 40bb54d | 2009-04-15 12:11:10 +0200 | [diff] [blame] | 3886 | spin_unlock_irq(q->queue_lock); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3887 | } |
| 3888 | |
| 3889 | /* |
| 3890 | * Timer running if the active_queue is currently idling inside its time slice |
| 3891 | */ |
| 3892 | static void cfq_idle_slice_timer(unsigned long data) |
| 3893 | { |
| 3894 | struct cfq_data *cfqd = (struct cfq_data *) data; |
| 3895 | struct cfq_queue *cfqq; |
| 3896 | unsigned long flags; |
Jens Axboe | 3c6bd2f | 2007-01-19 12:06:33 +1100 | [diff] [blame] | 3897 | int timed_out = 1; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3898 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3899 | cfq_log(cfqd, "idle timer fired"); |
| 3900 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3901 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
| 3902 | |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3903 | cfqq = cfqd->active_queue; |
| 3904 | if (cfqq) { |
Jens Axboe | 3c6bd2f | 2007-01-19 12:06:33 +1100 | [diff] [blame] | 3905 | timed_out = 0; |
| 3906 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3907 | /* |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 3908 | * We saw a request before the queue expired, let it through |
| 3909 | */ |
| 3910 | if (cfq_cfqq_must_dispatch(cfqq)) |
| 3911 | goto out_kick; |
| 3912 | |
| 3913 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3914 | * expired |
| 3915 | */ |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 3916 | if (cfq_slice_used(cfqq)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3917 | goto expire; |
| 3918 | |
| 3919 | /* |
| 3920 | * only expire and reinvoke request handler, if there are |
| 3921 | * other queues with pending requests |
| 3922 | */ |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3923 | if (!cfqd->busy_queues) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3924 | goto out_cont; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3925 | |
| 3926 | /* |
| 3927 | * not expired and it has a request pending, let it dispatch |
| 3928 | */ |
Jens Axboe | 75e5098 | 2009-04-07 08:56:14 +0200 | [diff] [blame] | 3929 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3930 | goto out_kick; |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3931 | |
| 3932 | /* |
| 3933 | * Queue depth flag is reset only when the idle didn't succeed |
| 3934 | */ |
| 3935 | cfq_clear_cfqq_deep(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3936 | } |
| 3937 | expire: |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3938 | cfq_slice_expired(cfqd, timed_out); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3939 | out_kick: |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3940 | cfq_schedule_dispatch(cfqd); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3941 | out_cont: |
| 3942 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
| 3943 | } |
| 3944 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3945 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
| 3946 | { |
| 3947 | del_timer_sync(&cfqd->idle_slice_timer); |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3948 | cancel_work_sync(&cfqd->unplug_work); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3949 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3950 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3951 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
| 3952 | { |
| 3953 | int i; |
| 3954 | |
| 3955 | for (i = 0; i < IOPRIO_BE_NR; i++) { |
| 3956 | if (cfqd->async_cfqq[0][i]) |
| 3957 | cfq_put_queue(cfqd->async_cfqq[0][i]); |
| 3958 | if (cfqd->async_cfqq[1][i]) |
| 3959 | cfq_put_queue(cfqd->async_cfqq[1][i]); |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3960 | } |
Oleg Nesterov | 2389d1e | 2007-11-05 08:58:05 +0100 | [diff] [blame] | 3961 | |
| 3962 | if (cfqd->async_idle_cfqq) |
| 3963 | cfq_put_queue(cfqd->async_idle_cfqq); |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3964 | } |
| 3965 | |
Jens Axboe | b374d18 | 2008-10-31 10:05:07 +0100 | [diff] [blame] | 3966 | static void cfq_exit_queue(struct elevator_queue *e) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3967 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3968 | struct cfq_data *cfqd = e->elevator_data; |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3969 | struct request_queue *q = cfqd->queue; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3970 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3971 | cfq_shutdown_timer_wq(cfqd); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3972 | |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 3973 | spin_lock_irq(q->queue_lock); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3974 | |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 3975 | if (cfqd->active_queue) |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3976 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3977 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3978 | cfq_put_async_queues(cfqd); |
Tejun Heo | 03aa264 | 2012-03-05 13:15:19 -0800 | [diff] [blame] | 3979 | |
| 3980 | spin_unlock_irq(q->queue_lock); |
| 3981 | |
Al Viro | a90d742 | 2006-03-18 12:05:37 -0500 | [diff] [blame] | 3982 | cfq_shutdown_timer_wq(cfqd); |
| 3983 | |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 3984 | #ifndef CONFIG_CFQ_GROUP_IOSCHED |
| 3985 | kfree(cfqd->root_group); |
Vivek Goyal | 2abae55 | 2011-05-23 10:02:19 +0200 | [diff] [blame] | 3986 | #endif |
Tejun Heo | e8989fa | 2012-03-05 13:15:20 -0800 | [diff] [blame] | 3987 | update_root_blkg_pd(q, BLKIO_POLICY_PROP); |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 3988 | kfree(cfqd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3989 | } |
| 3990 | |
Tejun Heo | b2fab5a | 2012-03-05 13:14:57 -0800 | [diff] [blame] | 3991 | static int cfq_init_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3992 | { |
| 3993 | struct cfq_data *cfqd; |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 3994 | struct blkio_group *blkg __maybe_unused; |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 3995 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3996 | |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 3997 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
Tejun Heo | a73f730 | 2011-12-14 00:33:37 +0100 | [diff] [blame] | 3998 | if (!cfqd) |
Tejun Heo | b2fab5a | 2012-03-05 13:14:57 -0800 | [diff] [blame] | 3999 | return -ENOMEM; |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 4000 | |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 4001 | cfqd->queue = q; |
| 4002 | q->elevator->elevator_data = cfqd; |
| 4003 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 4004 | /* Init root service tree */ |
| 4005 | cfqd->grp_service_tree = CFQ_RB_ROOT; |
| 4006 | |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 4007 | /* Init root group and prefer root group over other groups by default */ |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 4008 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 4009 | rcu_read_lock(); |
| 4010 | spin_lock_irq(q->queue_lock); |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 4011 | |
Tejun Heo | aaec55a | 2012-04-01 14:38:42 -0700 | [diff] [blame] | 4012 | blkg = blkg_lookup_create(&blkio_root_cgroup, q, true); |
Tejun Heo | cd1604f | 2012-03-05 13:15:06 -0800 | [diff] [blame] | 4013 | if (!IS_ERR(blkg)) |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 4014 | cfqd->root_group = blkg_to_cfqg(blkg); |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 4015 | |
| 4016 | spin_unlock_irq(q->queue_lock); |
| 4017 | rcu_read_unlock(); |
| 4018 | #else |
| 4019 | cfqd->root_group = kzalloc_node(sizeof(*cfqd->root_group), |
| 4020 | GFP_KERNEL, cfqd->queue->node); |
| 4021 | if (cfqd->root_group) |
| 4022 | cfq_init_cfqg_base(cfqd->root_group); |
| 4023 | #endif |
| 4024 | if (!cfqd->root_group) { |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 4025 | kfree(cfqd); |
Tejun Heo | b2fab5a | 2012-03-05 13:14:57 -0800 | [diff] [blame] | 4026 | return -ENOMEM; |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 4027 | } |
| 4028 | |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 4029 | cfqd->root_group->weight = 2*BLKIO_WEIGHT_DEFAULT; |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 4030 | |
Jens Axboe | 26a2ac0 | 2009-04-23 12:13:27 +0200 | [diff] [blame] | 4031 | /* |
| 4032 | * Not strictly needed (since RB_ROOT just clears the node and we |
| 4033 | * zeroed cfqd on alloc), but better be safe in case someone decides |
| 4034 | * to add magic to the rb code |
| 4035 | */ |
| 4036 | for (i = 0; i < CFQ_PRIO_LISTS; i++) |
| 4037 | cfqd->prio_trees[i] = RB_ROOT; |
| 4038 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 4039 | /* |
| 4040 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. |
| 4041 | * Grab a permanent reference to it, so that the normal code flow |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 4042 | * will not attempt to free it. oom_cfqq is linked to root_group |
| 4043 | * but shouldn't hold a reference as it'll never be unlinked. Lose |
| 4044 | * the reference from linking right away. |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 4045 | */ |
| 4046 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 4047 | cfqd->oom_cfqq.ref++; |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 4048 | |
| 4049 | spin_lock_irq(q->queue_lock); |
Tejun Heo | f51b802 | 2012-03-05 13:15:05 -0800 | [diff] [blame] | 4050 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, cfqd->root_group); |
Tejun Heo | eb7d8c07 | 2012-03-23 14:02:53 +0100 | [diff] [blame] | 4051 | cfqg_put(cfqd->root_group); |
Tejun Heo | 1adaf3d | 2012-03-05 13:15:15 -0800 | [diff] [blame] | 4052 | spin_unlock_irq(q->queue_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4053 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4054 | init_timer(&cfqd->idle_slice_timer); |
| 4055 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; |
| 4056 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; |
| 4057 | |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 4058 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4059 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4060 | cfqd->cfq_quantum = cfq_quantum; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4061 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
| 4062 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4063 | cfqd->cfq_back_max = cfq_back_max; |
| 4064 | cfqd->cfq_back_penalty = cfq_back_penalty; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4065 | cfqd->cfq_slice[0] = cfq_slice_async; |
| 4066 | cfqd->cfq_slice[1] = cfq_slice_sync; |
| 4067 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
| 4068 | cfqd->cfq_slice_idle = cfq_slice_idle; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4069 | cfqd->cfq_group_idle = cfq_group_idle; |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 4070 | cfqd->cfq_latency = 1; |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 4071 | cfqd->hw_tag = -1; |
Corrado Zoccolo | edc7113 | 2009-12-09 20:56:04 +0100 | [diff] [blame] | 4072 | /* |
| 4073 | * we optimistically start assuming sync ops weren't delayed in last |
| 4074 | * second, in order to have larger depth for async operations. |
| 4075 | */ |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 4076 | cfqd->last_delayed_sync = jiffies - HZ; |
Tejun Heo | b2fab5a | 2012-03-05 13:14:57 -0800 | [diff] [blame] | 4077 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4078 | } |
| 4079 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4080 | /* |
| 4081 | * sysfs parts below --> |
| 4082 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4083 | static ssize_t |
| 4084 | cfq_var_show(unsigned int var, char *page) |
| 4085 | { |
| 4086 | return sprintf(page, "%d\n", var); |
| 4087 | } |
| 4088 | |
| 4089 | static ssize_t |
| 4090 | cfq_var_store(unsigned int *var, const char *page, size_t count) |
| 4091 | { |
| 4092 | char *p = (char *) page; |
| 4093 | |
| 4094 | *var = simple_strtoul(p, &p, 10); |
| 4095 | return count; |
| 4096 | } |
| 4097 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4098 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
Jens Axboe | b374d18 | 2008-10-31 10:05:07 +0100 | [diff] [blame] | 4099 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4100 | { \ |
Al Viro | 3d1ab40 | 2006-03-18 18:35:43 -0500 | [diff] [blame] | 4101 | struct cfq_data *cfqd = e->elevator_data; \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4102 | unsigned int __data = __VAR; \ |
| 4103 | if (__CONV) \ |
| 4104 | __data = jiffies_to_msecs(__data); \ |
| 4105 | return cfq_var_show(__data, (page)); \ |
| 4106 | } |
| 4107 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4108 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
| 4109 | SHOW_FUNCTION(cfq_fifo_expire_async_show, cfqd->cfq_fifo_expire[0], 1); |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4110 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
| 4111 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4112 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4113 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4114 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
| 4115 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); |
| 4116 | SHOW_FUNCTION(cfq_slice_async_rq_show, cfqd->cfq_slice_async_rq, 0); |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 4117 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4118 | #undef SHOW_FUNCTION |
| 4119 | |
| 4120 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
Jens Axboe | b374d18 | 2008-10-31 10:05:07 +0100 | [diff] [blame] | 4121 | static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count) \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4122 | { \ |
Al Viro | 3d1ab40 | 2006-03-18 18:35:43 -0500 | [diff] [blame] | 4123 | struct cfq_data *cfqd = e->elevator_data; \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4124 | unsigned int __data; \ |
| 4125 | int ret = cfq_var_store(&__data, (page), count); \ |
| 4126 | if (__data < (MIN)) \ |
| 4127 | __data = (MIN); \ |
| 4128 | else if (__data > (MAX)) \ |
| 4129 | __data = (MAX); \ |
| 4130 | if (__CONV) \ |
| 4131 | *(__PTR) = msecs_to_jiffies(__data); \ |
| 4132 | else \ |
| 4133 | *(__PTR) = __data; \ |
| 4134 | return ret; \ |
| 4135 | } |
| 4136 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 4137 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
| 4138 | UINT_MAX, 1); |
| 4139 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, |
| 4140 | UINT_MAX, 1); |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4141 | STORE_FUNCTION(cfq_back_seek_max_store, &cfqd->cfq_back_max, 0, UINT_MAX, 0); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 4142 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
| 4143 | UINT_MAX, 0); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4144 | STORE_FUNCTION(cfq_slice_idle_store, &cfqd->cfq_slice_idle, 0, UINT_MAX, 1); |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4145 | STORE_FUNCTION(cfq_group_idle_store, &cfqd->cfq_group_idle, 0, UINT_MAX, 1); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4146 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
| 4147 | STORE_FUNCTION(cfq_slice_async_store, &cfqd->cfq_slice[0], 1, UINT_MAX, 1); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 4148 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
| 4149 | UINT_MAX, 0); |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 4150 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4151 | #undef STORE_FUNCTION |
| 4152 | |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4153 | #define CFQ_ATTR(name) \ |
| 4154 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 4155 | |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4156 | static struct elv_fs_entry cfq_attrs[] = { |
| 4157 | CFQ_ATTR(quantum), |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4158 | CFQ_ATTR(fifo_expire_sync), |
| 4159 | CFQ_ATTR(fifo_expire_async), |
| 4160 | CFQ_ATTR(back_seek_max), |
| 4161 | CFQ_ATTR(back_seek_penalty), |
| 4162 | CFQ_ATTR(slice_sync), |
| 4163 | CFQ_ATTR(slice_async), |
| 4164 | CFQ_ATTR(slice_async_rq), |
| 4165 | CFQ_ATTR(slice_idle), |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4166 | CFQ_ATTR(group_idle), |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 4167 | CFQ_ATTR(low_latency), |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4168 | __ATTR_NULL |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4169 | }; |
| 4170 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4171 | static struct elevator_type iosched_cfq = { |
| 4172 | .ops = { |
| 4173 | .elevator_merge_fn = cfq_merge, |
| 4174 | .elevator_merged_fn = cfq_merged_request, |
| 4175 | .elevator_merge_req_fn = cfq_merged_requests, |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 4176 | .elevator_allow_merge_fn = cfq_allow_merge, |
Divyesh Shah | 812d402 | 2010-04-08 21:14:23 -0700 | [diff] [blame] | 4177 | .elevator_bio_merged_fn = cfq_bio_merged, |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 4178 | .elevator_dispatch_fn = cfq_dispatch_requests, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4179 | .elevator_add_req_fn = cfq_insert_request, |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 4180 | .elevator_activate_req_fn = cfq_activate_request, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4181 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4182 | .elevator_completed_req_fn = cfq_completed_request, |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 4183 | .elevator_former_req_fn = elv_rb_former_request, |
| 4184 | .elevator_latter_req_fn = elv_rb_latter_request, |
Tejun Heo | 9b84cac | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4185 | .elevator_init_icq_fn = cfq_init_icq, |
Tejun Heo | 7e5a879 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4186 | .elevator_exit_icq_fn = cfq_exit_icq, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4187 | .elevator_set_req_fn = cfq_set_request, |
| 4188 | .elevator_put_req_fn = cfq_put_request, |
| 4189 | .elevator_may_queue_fn = cfq_may_queue, |
| 4190 | .elevator_init_fn = cfq_init_queue, |
| 4191 | .elevator_exit_fn = cfq_exit_queue, |
| 4192 | }, |
Tejun Heo | 3d3c237 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4193 | .icq_size = sizeof(struct cfq_io_cq), |
| 4194 | .icq_align = __alignof__(struct cfq_io_cq), |
Al Viro | 3d1ab40 | 2006-03-18 18:35:43 -0500 | [diff] [blame] | 4195 | .elevator_attrs = cfq_attrs, |
Tejun Heo | 3d3c237 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4196 | .elevator_name = "cfq", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4197 | .elevator_owner = THIS_MODULE, |
| 4198 | }; |
| 4199 | |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4200 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4201 | static struct blkio_policy_type blkio_policy_cfq = { |
| 4202 | .ops = { |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 4203 | .blkio_init_group_fn = cfq_init_blkio_group, |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4204 | }, |
Vivek Goyal | 062a644 | 2010-09-15 17:06:33 -0400 | [diff] [blame] | 4205 | .plid = BLKIO_POLICY_PROP, |
Tejun Heo | 0381411 | 2012-03-05 13:15:14 -0800 | [diff] [blame] | 4206 | .pdata_size = sizeof(struct cfq_group), |
Tejun Heo | 60c2bc2 | 2012-04-01 14:38:43 -0700 | [diff] [blame] | 4207 | .cftypes = cfq_blkcg_files, |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4208 | }; |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4209 | #endif |
| 4210 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4211 | static int __init cfq_init(void) |
| 4212 | { |
Tejun Heo | 3d3c237 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4213 | int ret; |
| 4214 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4215 | /* |
| 4216 | * could be 0 on HZ < 1000 setups |
| 4217 | */ |
| 4218 | if (!cfq_slice_async) |
| 4219 | cfq_slice_async = 1; |
| 4220 | if (!cfq_slice_idle) |
| 4221 | cfq_slice_idle = 1; |
| 4222 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4223 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4224 | if (!cfq_group_idle) |
| 4225 | cfq_group_idle = 1; |
| 4226 | #else |
| 4227 | cfq_group_idle = 0; |
| 4228 | #endif |
Tejun Heo | 3d3c237 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4229 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
| 4230 | if (!cfq_pool) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4231 | return -ENOMEM; |
| 4232 | |
Tejun Heo | 3d3c237 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4233 | ret = elv_register(&iosched_cfq); |
| 4234 | if (ret) { |
| 4235 | kmem_cache_destroy(cfq_pool); |
| 4236 | return ret; |
| 4237 | } |
Tejun Heo | 3d3c237 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4238 | |
Tejun Heo | b95ada5 | 2012-03-05 13:14:55 -0800 | [diff] [blame] | 4239 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4240 | blkio_policy_register(&blkio_policy_cfq); |
Tejun Heo | b95ada5 | 2012-03-05 13:14:55 -0800 | [diff] [blame] | 4241 | #endif |
Adrian Bunk | 2fdd82b | 2007-12-12 18:51:56 +0100 | [diff] [blame] | 4242 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4243 | } |
| 4244 | |
| 4245 | static void __exit cfq_exit(void) |
| 4246 | { |
Tejun Heo | b95ada5 | 2012-03-05 13:14:55 -0800 | [diff] [blame] | 4247 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4248 | blkio_policy_unregister(&blkio_policy_cfq); |
Tejun Heo | b95ada5 | 2012-03-05 13:14:55 -0800 | [diff] [blame] | 4249 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4250 | elv_unregister(&iosched_cfq); |
Tejun Heo | 3d3c237 | 2011-12-14 00:33:42 +0100 | [diff] [blame] | 4251 | kmem_cache_destroy(cfq_pool); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4252 | } |
| 4253 | |
| 4254 | module_init(cfq_init); |
| 4255 | module_exit(cfq_exit); |
| 4256 | |
| 4257 | MODULE_AUTHOR("Jens Axboe"); |
| 4258 | MODULE_LICENSE("GPL"); |
| 4259 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |