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> |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 17 | #include "cfq.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | |
| 19 | /* |
| 20 | * tunables |
| 21 | */ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 22 | /* max queue in one round of service */ |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 23 | static const int cfq_quantum = 8; |
Arjan van de Ven | 6410009 | 2006-01-06 09:46:02 +0100 | [diff] [blame] | 24 | static const int cfq_fifo_expire[2] = { HZ / 4, HZ / 8 }; |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 25 | /* maximum backwards seek, in KiB */ |
| 26 | static const int cfq_back_max = 16 * 1024; |
| 27 | /* penalty of a backwards seek */ |
| 28 | static const int cfq_back_penalty = 2; |
Arjan van de Ven | 6410009 | 2006-01-06 09:46:02 +0100 | [diff] [blame] | 29 | static const int cfq_slice_sync = HZ / 10; |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 30 | static int cfq_slice_async = HZ / 25; |
Arjan van de Ven | 6410009 | 2006-01-06 09:46:02 +0100 | [diff] [blame] | 31 | static const int cfq_slice_async_rq = 2; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 32 | static int cfq_slice_idle = HZ / 125; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 33 | static int cfq_group_idle = HZ / 125; |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 34 | static const int cfq_target_latency = HZ * 3/10; /* 300 ms */ |
| 35 | static const int cfq_hist_divisor = 4; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 36 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 37 | /* |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 38 | * offset from end of service tree |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 39 | */ |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 40 | #define CFQ_IDLE_DELAY (HZ / 5) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 41 | |
| 42 | /* |
| 43 | * below this threshold, we consider thinktime immediate |
| 44 | */ |
| 45 | #define CFQ_MIN_TT (2) |
| 46 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 47 | #define CFQ_SLICE_SCALE (5) |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 48 | #define CFQ_HW_QUEUE_MIN (5) |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 49 | #define CFQ_SERVICE_SHIFT 12 |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 50 | |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 51 | #define CFQQ_SEEK_THR (sector_t)(8 * 100) |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 52 | #define CFQQ_CLOSE_THR (sector_t)(8 * 1024) |
Corrado Zoccolo | 41647e7 | 2010-02-27 19:45:40 +0100 | [diff] [blame] | 53 | #define CFQQ_SECT_THR_NONROT (sector_t)(2 * 32) |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 54 | #define CFQQ_SEEKY(cfqq) (hweight32(cfqq->seek_history) > 32/8) |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 55 | |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 56 | #define RQ_CIC(rq) \ |
Mike Snitzer | c186794 | 2011-02-11 11:08:00 +0100 | [diff] [blame] | 57 | ((struct cfq_io_context *) (rq)->elevator_private[0]) |
| 58 | #define RQ_CFQQ(rq) (struct cfq_queue *) ((rq)->elevator_private[1]) |
| 59 | #define RQ_CFQG(rq) (struct cfq_group *) ((rq)->elevator_private[2]) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | |
Christoph Lameter | e18b890 | 2006-12-06 20:33:20 -0800 | [diff] [blame] | 61 | static struct kmem_cache *cfq_pool; |
| 62 | static struct kmem_cache *cfq_ioc_pool; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 63 | |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 64 | static DEFINE_PER_CPU(unsigned long, cfq_ioc_count); |
Al Viro | 334e94d | 2006-03-18 15:05:53 -0500 | [diff] [blame] | 65 | static struct completion *ioc_gone; |
Jens Axboe | 9a11b4e | 2008-05-29 09:32:08 +0200 | [diff] [blame] | 66 | static DEFINE_SPINLOCK(ioc_gone_lock); |
Al Viro | 334e94d | 2006-03-18 15:05:53 -0500 | [diff] [blame] | 67 | |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 68 | static DEFINE_SPINLOCK(cic_index_lock); |
| 69 | static DEFINE_IDA(cic_index_ida); |
| 70 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 71 | #define CFQ_PRIO_LISTS IOPRIO_BE_NR |
| 72 | #define cfq_class_idle(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_IDLE) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 73 | #define cfq_class_rt(cfqq) ((cfqq)->ioprio_class == IOPRIO_CLASS_RT) |
| 74 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 75 | #define sample_valid(samples) ((samples) > 80) |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 76 | #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] | 77 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 78 | /* |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 79 | * Most of our rbtree usage is for sorting with min extraction, so |
| 80 | * if we cache the leftmost node we don't have to walk down the tree |
| 81 | * to find it. Idea borrowed from Ingo Molnars CFS scheduler. We should |
| 82 | * move this into the elevator for the rq sorting as well. |
| 83 | */ |
| 84 | struct cfq_rb_root { |
| 85 | struct rb_root rb; |
| 86 | struct rb_node *left; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 87 | unsigned count; |
Richard Kennedy | 73e9ffd | 2010-03-01 10:50:20 +0100 | [diff] [blame] | 88 | unsigned total_weight; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 89 | u64 min_vdisktime; |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 90 | struct cfq_ttime ttime; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 91 | }; |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 92 | #define CFQ_RB_ROOT (struct cfq_rb_root) { .rb = RB_ROOT, \ |
| 93 | .ttime = {.last_end_request = jiffies,},} |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 94 | |
| 95 | /* |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 96 | * Per process-grouping structure |
| 97 | */ |
| 98 | struct cfq_queue { |
| 99 | /* reference count */ |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 100 | int ref; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 101 | /* various state flags, see below */ |
| 102 | unsigned int flags; |
| 103 | /* parent cfq_data */ |
| 104 | struct cfq_data *cfqd; |
| 105 | /* service_tree member */ |
| 106 | struct rb_node rb_node; |
| 107 | /* service_tree key */ |
| 108 | unsigned long rb_key; |
| 109 | /* prio tree member */ |
| 110 | struct rb_node p_node; |
| 111 | /* prio tree root we belong to, if any */ |
| 112 | struct rb_root *p_root; |
| 113 | /* sorted list of pending requests */ |
| 114 | struct rb_root sort_list; |
| 115 | /* if fifo isn't expired, next request to serve */ |
| 116 | struct request *next_rq; |
| 117 | /* requests queued in sort_list */ |
| 118 | int queued[2]; |
| 119 | /* currently allocated requests */ |
| 120 | int allocated[2]; |
| 121 | /* fifo list of requests in sort_list */ |
| 122 | struct list_head fifo; |
| 123 | |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 124 | /* time when queue got scheduled in to dispatch first request. */ |
| 125 | unsigned long dispatch_start; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 126 | unsigned int allocated_slice; |
Richard Kennedy | c4081ba | 2010-02-22 13:49:24 +0100 | [diff] [blame] | 127 | unsigned int slice_dispatch; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 128 | /* time when first request from queue completed and slice started. */ |
| 129 | unsigned long slice_start; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 130 | unsigned long slice_end; |
| 131 | long slice_resid; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 132 | |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 133 | /* pending priority requests */ |
| 134 | int prio_pending; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 135 | /* number of requests that are on the dispatch list or inside driver */ |
| 136 | int dispatched; |
| 137 | |
| 138 | /* io prio of this group */ |
| 139 | unsigned short ioprio, org_ioprio; |
Justin TerAvest | 4aede84 | 2011-07-12 08:31:45 +0200 | [diff] [blame] | 140 | unsigned short ioprio_class; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 141 | |
Richard Kennedy | c4081ba | 2010-02-22 13:49:24 +0100 | [diff] [blame] | 142 | pid_t pid; |
| 143 | |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 144 | u32 seek_history; |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 145 | sector_t last_request_pos; |
| 146 | |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 147 | struct cfq_rb_root *service_tree; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 148 | struct cfq_queue *new_cfqq; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 149 | struct cfq_group *cfqg; |
Vivek Goyal | c4e7893 | 2010-08-23 12:25:03 +0200 | [diff] [blame] | 150 | /* Number of sectors dispatched from queue in single dispatch round */ |
| 151 | unsigned long nr_sectors; |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 152 | }; |
| 153 | |
| 154 | /* |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 155 | * First index in the service_trees. |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 156 | * IDLE is handled separately, so it has negative index |
| 157 | */ |
| 158 | enum wl_prio_t { |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 159 | BE_WORKLOAD = 0, |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 160 | RT_WORKLOAD = 1, |
| 161 | IDLE_WORKLOAD = 2, |
Vivek Goyal | b462732 | 2010-10-22 09:48:43 +0200 | [diff] [blame] | 162 | CFQ_PRIO_NR, |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 163 | }; |
| 164 | |
| 165 | /* |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 166 | * Second index in the service_trees. |
| 167 | */ |
| 168 | enum wl_type_t { |
| 169 | ASYNC_WORKLOAD = 0, |
| 170 | SYNC_NOIDLE_WORKLOAD = 1, |
| 171 | SYNC_WORKLOAD = 2 |
| 172 | }; |
| 173 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 174 | /* This is per cgroup per device grouping structure */ |
| 175 | struct cfq_group { |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 176 | /* group service_tree member */ |
| 177 | struct rb_node rb_node; |
| 178 | |
| 179 | /* group service_tree key */ |
| 180 | u64 vdisktime; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 181 | unsigned int weight; |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 182 | unsigned int new_weight; |
| 183 | bool needs_update; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 184 | |
| 185 | /* number of cfqq currently on this group */ |
| 186 | int nr_cfqq; |
| 187 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 188 | /* |
Kyungmin Park | 4495a7d | 2011-05-31 10:04:09 +0200 | [diff] [blame] | 189 | * Per group busy queues average. Useful for workload slice calc. We |
Vivek Goyal | b462732 | 2010-10-22 09:48:43 +0200 | [diff] [blame] | 190 | * create the array for each prio class but at run time it is used |
| 191 | * only for RT and BE class and slot for IDLE class remains unused. |
| 192 | * This is primarily done to avoid confusion and a gcc warning. |
| 193 | */ |
| 194 | unsigned int busy_queues_avg[CFQ_PRIO_NR]; |
| 195 | /* |
| 196 | * rr lists of queues with requests. We maintain service trees for |
| 197 | * RT and BE classes. These trees are subdivided in subclasses |
| 198 | * of SYNC, SYNC_NOIDLE and ASYNC based on workload type. For IDLE |
| 199 | * class there is no subclassification and all the cfq queues go on |
| 200 | * a single tree service_tree_idle. |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 201 | * Counts are embedded in the cfq_rb_root |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 202 | */ |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 203 | struct cfq_rb_root service_trees[2][3]; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 204 | struct cfq_rb_root service_tree_idle; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 205 | |
| 206 | unsigned long saved_workload_slice; |
| 207 | enum wl_type_t saved_workload; |
| 208 | enum wl_prio_t saved_serving_prio; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 209 | struct blkio_group blkg; |
| 210 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 211 | struct hlist_node cfqd_node; |
Shaohua Li | 329a678 | 2011-01-07 08:48:28 +0100 | [diff] [blame] | 212 | int ref; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 213 | #endif |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 214 | /* number of requests that are on the dispatch list or inside driver */ |
| 215 | int dispatched; |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 216 | struct cfq_ttime ttime; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 217 | }; |
| 218 | |
| 219 | /* |
| 220 | * Per block device queue structure |
| 221 | */ |
| 222 | struct cfq_data { |
| 223 | struct request_queue *queue; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 224 | /* Root service tree for cfq_groups */ |
| 225 | struct cfq_rb_root grp_service_tree; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 226 | struct cfq_group root_group; |
| 227 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 228 | /* |
| 229 | * The priority currently being served |
| 230 | */ |
| 231 | enum wl_prio_t serving_prio; |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 232 | enum wl_type_t serving_type; |
| 233 | unsigned long workload_expires; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 234 | struct cfq_group *serving_group; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 235 | |
| 236 | /* |
| 237 | * Each priority tree is sorted by next_request position. These |
| 238 | * trees are used when determining if two or more queues are |
| 239 | * interleaving requests (see cfq_close_cooperator). |
| 240 | */ |
| 241 | struct rb_root prio_trees[CFQ_PRIO_LISTS]; |
| 242 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 243 | unsigned int busy_queues; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 244 | unsigned int busy_sync_queues; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 245 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 246 | int rq_in_driver; |
| 247 | int rq_in_flight[2]; |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 248 | |
| 249 | /* |
| 250 | * queue-depth detection |
| 251 | */ |
| 252 | int rq_queued; |
Jens Axboe | 25776e3 | 2006-06-01 10:12:26 +0200 | [diff] [blame] | 253 | int hw_tag; |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 254 | /* |
| 255 | * hw_tag can be |
| 256 | * -1 => indeterminate, (cfq will behave as if NCQ is present, to allow better detection) |
| 257 | * 1 => NCQ is present (hw_tag_est_depth is the estimated max depth) |
| 258 | * 0 => no NCQ |
| 259 | */ |
| 260 | int hw_tag_est_depth; |
| 261 | unsigned int hw_tag_samples; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 262 | |
| 263 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 264 | * idle window management |
| 265 | */ |
| 266 | struct timer_list idle_slice_timer; |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 267 | struct work_struct unplug_work; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 268 | |
| 269 | struct cfq_queue *active_queue; |
| 270 | struct cfq_io_context *active_cic; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 271 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 272 | /* |
| 273 | * async queue for each priority case |
| 274 | */ |
| 275 | struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR]; |
| 276 | struct cfq_queue *async_idle_cfqq; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 277 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 278 | sector_t last_position; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 279 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | /* |
| 281 | * tunables, see top of file |
| 282 | */ |
| 283 | unsigned int cfq_quantum; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 284 | unsigned int cfq_fifo_expire[2]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 285 | unsigned int cfq_back_penalty; |
| 286 | unsigned int cfq_back_max; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 287 | unsigned int cfq_slice[2]; |
| 288 | unsigned int cfq_slice_async_rq; |
| 289 | unsigned int cfq_slice_idle; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 290 | unsigned int cfq_group_idle; |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 291 | unsigned int cfq_latency; |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 292 | |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 293 | unsigned int cic_index; |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 294 | struct list_head cic_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 296 | /* |
| 297 | * Fallback dummy cfqq for extreme OOM conditions |
| 298 | */ |
| 299 | struct cfq_queue oom_cfqq; |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 300 | |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 301 | unsigned long last_delayed_sync; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 302 | |
| 303 | /* List of cfq groups being managed on this device*/ |
| 304 | struct hlist_head cfqg_list; |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 305 | |
| 306 | /* Number of groups which are on blkcg->blkg_list */ |
| 307 | unsigned int nr_blkcg_linked_grps; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | }; |
| 309 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 310 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd); |
| 311 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 312 | static struct cfq_rb_root *service_tree_for(struct cfq_group *cfqg, |
| 313 | enum wl_prio_t prio, |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 314 | enum wl_type_t type) |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 315 | { |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 316 | if (!cfqg) |
| 317 | return NULL; |
| 318 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 319 | if (prio == IDLE_WORKLOAD) |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 320 | return &cfqg->service_tree_idle; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 321 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 322 | return &cfqg->service_trees[prio][type]; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 323 | } |
| 324 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 325 | enum cfqq_state_flags { |
Jens Axboe | b0b8d749 | 2007-01-19 11:35:30 +1100 | [diff] [blame] | 326 | CFQ_CFQQ_FLAG_on_rr = 0, /* on round-robin busy list */ |
| 327 | CFQ_CFQQ_FLAG_wait_request, /* waiting for a request */ |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 328 | CFQ_CFQQ_FLAG_must_dispatch, /* must be allowed a dispatch */ |
Jens Axboe | b0b8d749 | 2007-01-19 11:35:30 +1100 | [diff] [blame] | 329 | CFQ_CFQQ_FLAG_must_alloc_slice, /* per-slice must_alloc flag */ |
Jens Axboe | b0b8d749 | 2007-01-19 11:35:30 +1100 | [diff] [blame] | 330 | CFQ_CFQQ_FLAG_fifo_expire, /* FIFO checked in this slice */ |
| 331 | CFQ_CFQQ_FLAG_idle_window, /* slice idling enabled */ |
| 332 | CFQ_CFQQ_FLAG_prio_changed, /* task priority has changed */ |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 333 | CFQ_CFQQ_FLAG_slice_new, /* no requests dispatched in slice */ |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 334 | CFQ_CFQQ_FLAG_sync, /* synchronous queue */ |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 335 | CFQ_CFQQ_FLAG_coop, /* cfqq is shared */ |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 336 | CFQ_CFQQ_FLAG_split_coop, /* shared cfqq will be splitted */ |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 337 | CFQ_CFQQ_FLAG_deep, /* sync cfqq experienced large depth */ |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 338 | CFQ_CFQQ_FLAG_wait_busy, /* Waiting for next request */ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 339 | }; |
| 340 | |
| 341 | #define CFQ_CFQQ_FNS(name) \ |
| 342 | static inline void cfq_mark_cfqq_##name(struct cfq_queue *cfqq) \ |
| 343 | { \ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 344 | (cfqq)->flags |= (1 << CFQ_CFQQ_FLAG_##name); \ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 345 | } \ |
| 346 | static inline void cfq_clear_cfqq_##name(struct cfq_queue *cfqq) \ |
| 347 | { \ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 348 | (cfqq)->flags &= ~(1 << CFQ_CFQQ_FLAG_##name); \ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 349 | } \ |
| 350 | static inline int cfq_cfqq_##name(const struct cfq_queue *cfqq) \ |
| 351 | { \ |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 352 | return ((cfqq)->flags & (1 << CFQ_CFQQ_FLAG_##name)) != 0; \ |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 353 | } |
| 354 | |
| 355 | CFQ_CFQQ_FNS(on_rr); |
| 356 | CFQ_CFQQ_FNS(wait_request); |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 357 | CFQ_CFQQ_FNS(must_dispatch); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 358 | CFQ_CFQQ_FNS(must_alloc_slice); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 359 | CFQ_CFQQ_FNS(fifo_expire); |
| 360 | CFQ_CFQQ_FNS(idle_window); |
| 361 | CFQ_CFQQ_FNS(prio_changed); |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 362 | CFQ_CFQQ_FNS(slice_new); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 363 | CFQ_CFQQ_FNS(sync); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 364 | CFQ_CFQQ_FNS(coop); |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 365 | CFQ_CFQQ_FNS(split_coop); |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 366 | CFQ_CFQQ_FNS(deep); |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 367 | CFQ_CFQQ_FNS(wait_busy); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 368 | #undef CFQ_CFQQ_FNS |
| 369 | |
Vivek Goyal | afc24d4 | 2010-04-26 19:27:56 +0200 | [diff] [blame] | 370 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 371 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
| 372 | blk_add_trace_msg((cfqd)->queue, "cfq%d%c %s " fmt, (cfqq)->pid, \ |
| 373 | cfq_cfqq_sync((cfqq)) ? 'S' : 'A', \ |
Kyungmin Park | 4495a7d | 2011-05-31 10:04:09 +0200 | [diff] [blame] | 374 | blkg_path(&(cfqq)->cfqg->blkg), ##args) |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 375 | |
| 376 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) \ |
| 377 | blk_add_trace_msg((cfqd)->queue, "%s " fmt, \ |
Kyungmin Park | 4495a7d | 2011-05-31 10:04:09 +0200 | [diff] [blame] | 378 | blkg_path(&(cfqg)->blkg), ##args) \ |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 379 | |
| 380 | #else |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 381 | #define cfq_log_cfqq(cfqd, cfqq, fmt, args...) \ |
| 382 | blk_add_trace_msg((cfqd)->queue, "cfq%d " fmt, (cfqq)->pid, ##args) |
Kyungmin Park | 4495a7d | 2011-05-31 10:04:09 +0200 | [diff] [blame] | 383 | #define cfq_log_cfqg(cfqd, cfqg, fmt, args...) do {} while (0) |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 384 | #endif |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 385 | #define cfq_log(cfqd, fmt, args...) \ |
| 386 | blk_add_trace_msg((cfqd)->queue, "cfq " fmt, ##args) |
| 387 | |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 388 | /* Traverses through cfq group service trees */ |
| 389 | #define for_each_cfqg_st(cfqg, i, j, st) \ |
| 390 | for (i = 0; i <= IDLE_WORKLOAD; i++) \ |
| 391 | for (j = 0, st = i < IDLE_WORKLOAD ? &cfqg->service_trees[i][j]\ |
| 392 | : &cfqg->service_tree_idle; \ |
| 393 | (i < IDLE_WORKLOAD && j <= SYNC_WORKLOAD) || \ |
| 394 | (i == IDLE_WORKLOAD && j == 0); \ |
| 395 | j++, st = i < IDLE_WORKLOAD ? \ |
| 396 | &cfqg->service_trees[i][j]: NULL) \ |
| 397 | |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 398 | static inline bool cfq_io_thinktime_big(struct cfq_data *cfqd, |
| 399 | struct cfq_ttime *ttime, bool group_idle) |
| 400 | { |
| 401 | unsigned long slice; |
| 402 | if (!sample_valid(ttime->ttime_samples)) |
| 403 | return false; |
| 404 | if (group_idle) |
| 405 | slice = cfqd->cfq_group_idle; |
| 406 | else |
| 407 | slice = cfqd->cfq_slice_idle; |
| 408 | return ttime->ttime_mean > slice; |
| 409 | } |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 410 | |
Vivek Goyal | 02b3508 | 2010-08-23 12:23:53 +0200 | [diff] [blame] | 411 | static inline bool iops_mode(struct cfq_data *cfqd) |
| 412 | { |
| 413 | /* |
| 414 | * If we are not idling on queues and it is a NCQ drive, parallel |
| 415 | * execution of requests is on and measuring time is not possible |
| 416 | * in most of the cases until and unless we drive shallower queue |
| 417 | * depths and that becomes a performance bottleneck. In such cases |
| 418 | * switch to start providing fairness in terms of number of IOs. |
| 419 | */ |
| 420 | if (!cfqd->cfq_slice_idle && cfqd->hw_tag) |
| 421 | return true; |
| 422 | else |
| 423 | return false; |
| 424 | } |
| 425 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 426 | static inline enum wl_prio_t cfqq_prio(struct cfq_queue *cfqq) |
| 427 | { |
| 428 | if (cfq_class_idle(cfqq)) |
| 429 | return IDLE_WORKLOAD; |
| 430 | if (cfq_class_rt(cfqq)) |
| 431 | return RT_WORKLOAD; |
| 432 | return BE_WORKLOAD; |
| 433 | } |
| 434 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 435 | |
| 436 | static enum wl_type_t cfqq_type(struct cfq_queue *cfqq) |
| 437 | { |
| 438 | if (!cfq_cfqq_sync(cfqq)) |
| 439 | return ASYNC_WORKLOAD; |
| 440 | if (!cfq_cfqq_idle_window(cfqq)) |
| 441 | return SYNC_NOIDLE_WORKLOAD; |
| 442 | return SYNC_WORKLOAD; |
| 443 | } |
| 444 | |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 445 | static inline int cfq_group_busy_queues_wl(enum wl_prio_t wl, |
| 446 | struct cfq_data *cfqd, |
| 447 | struct cfq_group *cfqg) |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 448 | { |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 449 | if (wl == IDLE_WORKLOAD) |
| 450 | return cfqg->service_tree_idle.count; |
| 451 | |
| 452 | return cfqg->service_trees[wl][ASYNC_WORKLOAD].count |
| 453 | + cfqg->service_trees[wl][SYNC_NOIDLE_WORKLOAD].count |
| 454 | + cfqg->service_trees[wl][SYNC_WORKLOAD].count; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 455 | } |
| 456 | |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 457 | static inline int cfqg_busy_async_queues(struct cfq_data *cfqd, |
| 458 | struct cfq_group *cfqg) |
| 459 | { |
| 460 | return cfqg->service_trees[RT_WORKLOAD][ASYNC_WORKLOAD].count |
| 461 | + cfqg->service_trees[BE_WORKLOAD][ASYNC_WORKLOAD].count; |
| 462 | } |
| 463 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 464 | static void cfq_dispatch_insert(struct request_queue *, struct request *); |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 465 | static struct cfq_queue *cfq_get_queue(struct cfq_data *, bool, |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 466 | struct io_context *, gfp_t); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 467 | static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *, |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 468 | struct io_context *); |
| 469 | |
| 470 | static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic, |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 471 | bool is_sync) |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 472 | { |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 473 | return cic->cfqq[is_sync]; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 474 | } |
| 475 | |
| 476 | static inline void cic_set_cfqq(struct cfq_io_context *cic, |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 477 | struct cfq_queue *cfqq, bool is_sync) |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 478 | { |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 479 | cic->cfqq[is_sync] = cfqq; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 480 | } |
| 481 | |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 482 | #define CIC_DEAD_KEY 1ul |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 483 | #define CIC_DEAD_INDEX_SHIFT 1 |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 484 | |
| 485 | static inline void *cfqd_dead_key(struct cfq_data *cfqd) |
| 486 | { |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 487 | return (void *)(cfqd->cic_index << CIC_DEAD_INDEX_SHIFT | CIC_DEAD_KEY); |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 488 | } |
| 489 | |
| 490 | static inline struct cfq_data *cic_to_cfqd(struct cfq_io_context *cic) |
| 491 | { |
| 492 | struct cfq_data *cfqd = cic->key; |
| 493 | |
| 494 | if (unlikely((unsigned long) cfqd & CIC_DEAD_KEY)) |
| 495 | return NULL; |
| 496 | |
| 497 | return cfqd; |
| 498 | } |
| 499 | |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 500 | /* |
| 501 | * We regard a request as SYNC, if it's either a read or has the SYNC bit |
| 502 | * set (in which case it could also be direct WRITE). |
| 503 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 504 | static inline bool cfq_bio_sync(struct bio *bio) |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 505 | { |
Christoph Hellwig | 7b6d91d | 2010-08-07 18:20:39 +0200 | [diff] [blame] | 506 | return bio_data_dir(bio) == READ || (bio->bi_rw & REQ_SYNC); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 507 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 508 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | /* |
Andrew Morton | 99f95e5 | 2005-06-27 20:14:05 -0700 | [diff] [blame] | 510 | * scheduler run of queue, if there are requests pending and no one in the |
| 511 | * driver that will restart queueing |
| 512 | */ |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 513 | static inline void cfq_schedule_dispatch(struct cfq_data *cfqd) |
Andrew Morton | 99f95e5 | 2005-06-27 20:14:05 -0700 | [diff] [blame] | 514 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 515 | if (cfqd->busy_queues) { |
| 516 | cfq_log(cfqd, "schedule dispatch"); |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 517 | kblockd_schedule_work(cfqd->queue, &cfqd->unplug_work); |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 518 | } |
Andrew Morton | 99f95e5 | 2005-06-27 20:14:05 -0700 | [diff] [blame] | 519 | } |
| 520 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | /* |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 522 | * Scale schedule slice based on io priority. Use the sync time slice only |
| 523 | * if a queue is marked sync and has sync io queued. A sync queue with async |
| 524 | * io only, should not get full sync slice length. |
| 525 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 526 | static inline int cfq_prio_slice(struct cfq_data *cfqd, bool sync, |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 527 | unsigned short prio) |
| 528 | { |
| 529 | const int base_slice = cfqd->cfq_slice[sync]; |
| 530 | |
| 531 | WARN_ON(prio >= IOPRIO_BE_NR); |
| 532 | |
| 533 | return base_slice + (base_slice/CFQ_SLICE_SCALE * (4 - prio)); |
| 534 | } |
| 535 | |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 536 | static inline int |
| 537 | cfq_prio_to_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 538 | { |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 539 | return cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio); |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 540 | } |
| 541 | |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 542 | static inline u64 cfq_scale_slice(unsigned long delta, struct cfq_group *cfqg) |
| 543 | { |
| 544 | u64 d = delta << CFQ_SERVICE_SHIFT; |
| 545 | |
| 546 | d = d * BLKIO_WEIGHT_DEFAULT; |
| 547 | do_div(d, cfqg->weight); |
| 548 | return d; |
| 549 | } |
| 550 | |
| 551 | static inline u64 max_vdisktime(u64 min_vdisktime, u64 vdisktime) |
| 552 | { |
| 553 | s64 delta = (s64)(vdisktime - min_vdisktime); |
| 554 | if (delta > 0) |
| 555 | min_vdisktime = vdisktime; |
| 556 | |
| 557 | return min_vdisktime; |
| 558 | } |
| 559 | |
| 560 | static inline u64 min_vdisktime(u64 min_vdisktime, u64 vdisktime) |
| 561 | { |
| 562 | s64 delta = (s64)(vdisktime - min_vdisktime); |
| 563 | if (delta < 0) |
| 564 | min_vdisktime = vdisktime; |
| 565 | |
| 566 | return min_vdisktime; |
| 567 | } |
| 568 | |
| 569 | static void update_min_vdisktime(struct cfq_rb_root *st) |
| 570 | { |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 571 | struct cfq_group *cfqg; |
| 572 | |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 573 | if (st->left) { |
| 574 | cfqg = rb_entry_cfqg(st->left); |
Gui Jianfeng | a603271 | 2011-03-07 09:28:09 +0100 | [diff] [blame] | 575 | st->min_vdisktime = max_vdisktime(st->min_vdisktime, |
| 576 | cfqg->vdisktime); |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 577 | } |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 578 | } |
| 579 | |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 580 | /* |
| 581 | * get averaged number of queues of RT/BE priority. |
| 582 | * average is updated, with a formula that gives more weight to higher numbers, |
| 583 | * to quickly follows sudden increases and decrease slowly |
| 584 | */ |
| 585 | |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 586 | static inline unsigned cfq_group_get_avg_queues(struct cfq_data *cfqd, |
| 587 | struct cfq_group *cfqg, bool rt) |
Jens Axboe | 5869619c | 2009-10-28 09:27:07 +0100 | [diff] [blame] | 588 | { |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 589 | unsigned min_q, max_q; |
| 590 | unsigned mult = cfq_hist_divisor - 1; |
| 591 | unsigned round = cfq_hist_divisor / 2; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 592 | unsigned busy = cfq_group_busy_queues_wl(rt, cfqd, cfqg); |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 593 | |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 594 | min_q = min(cfqg->busy_queues_avg[rt], busy); |
| 595 | max_q = max(cfqg->busy_queues_avg[rt], busy); |
| 596 | cfqg->busy_queues_avg[rt] = (mult * max_q + min_q + round) / |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 597 | cfq_hist_divisor; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 598 | return cfqg->busy_queues_avg[rt]; |
| 599 | } |
| 600 | |
| 601 | static inline unsigned |
| 602 | cfq_group_slice(struct cfq_data *cfqd, struct cfq_group *cfqg) |
| 603 | { |
| 604 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 605 | |
| 606 | return cfq_target_latency * cfqg->weight / st->total_weight; |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 607 | } |
| 608 | |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 609 | static inline unsigned |
Vivek Goyal | ba5bd52 | 2011-01-19 08:25:02 -0700 | [diff] [blame] | 610 | cfq_scaled_cfqq_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 611 | { |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 612 | unsigned slice = cfq_prio_to_slice(cfqd, cfqq); |
| 613 | if (cfqd->cfq_latency) { |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 614 | /* |
| 615 | * interested queues (we consider only the ones with the same |
| 616 | * priority class in the cfq group) |
| 617 | */ |
| 618 | unsigned iq = cfq_group_get_avg_queues(cfqd, cfqq->cfqg, |
| 619 | cfq_class_rt(cfqq)); |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 620 | unsigned sync_slice = cfqd->cfq_slice[1]; |
| 621 | unsigned expect_latency = sync_slice * iq; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 622 | unsigned group_slice = cfq_group_slice(cfqd, cfqq->cfqg); |
| 623 | |
| 624 | if (expect_latency > group_slice) { |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 625 | unsigned base_low_slice = 2 * cfqd->cfq_slice_idle; |
| 626 | /* scale low_slice according to IO priority |
| 627 | * and sync vs async */ |
| 628 | unsigned low_slice = |
| 629 | min(slice, base_low_slice * slice / sync_slice); |
| 630 | /* the adapted slice value is scaled to fit all iqs |
| 631 | * into the target latency */ |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 632 | slice = max(slice * group_slice / expect_latency, |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 633 | low_slice); |
| 634 | } |
| 635 | } |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 636 | return slice; |
| 637 | } |
| 638 | |
| 639 | static inline void |
| 640 | cfq_set_prio_slice(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 641 | { |
Vivek Goyal | ba5bd52 | 2011-01-19 08:25:02 -0700 | [diff] [blame] | 642 | unsigned slice = cfq_scaled_cfqq_slice(cfqd, cfqq); |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 643 | |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 644 | cfqq->slice_start = jiffies; |
Corrado Zoccolo | 5db5d64 | 2009-10-26 22:44:04 +0100 | [diff] [blame] | 645 | cfqq->slice_end = jiffies + slice; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 646 | cfqq->allocated_slice = slice; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 647 | cfq_log_cfqq(cfqd, cfqq, "set_slice=%lu", cfqq->slice_end - jiffies); |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 648 | } |
| 649 | |
| 650 | /* |
| 651 | * We need to wrap this check in cfq_cfqq_slice_new(), since ->slice_end |
| 652 | * isn't valid until the first request from the dispatch is activated |
| 653 | * and the slice time set. |
| 654 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 655 | static inline bool cfq_slice_used(struct cfq_queue *cfqq) |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 656 | { |
| 657 | if (cfq_cfqq_slice_new(cfqq)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 658 | return false; |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 659 | if (time_before(jiffies, cfqq->slice_end)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 660 | return false; |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 661 | |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 662 | return true; |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 663 | } |
| 664 | |
| 665 | /* |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 666 | * 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] | 667 | * 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] | 668 | * behind the head is penalized and only allowed to a certain extent. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | */ |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 670 | static struct request * |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 671 | 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] | 672 | { |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 673 | sector_t s1, s2, d1 = 0, d2 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | unsigned long back_max; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 675 | #define CFQ_RQ1_WRAP 0x01 /* request 1 wraps */ |
| 676 | #define CFQ_RQ2_WRAP 0x02 /* request 2 wraps */ |
| 677 | unsigned wrap = 0; /* bit mask: requests behind the disk head? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 679 | if (rq1 == NULL || rq1 == rq2) |
| 680 | return rq2; |
| 681 | if (rq2 == NULL) |
| 682 | return rq1; |
Jens Axboe | 9c2c38a | 2005-08-24 14:57:54 +0200 | [diff] [blame] | 683 | |
Namhyung Kim | 229836b | 2011-05-24 10:23:21 +0200 | [diff] [blame] | 684 | if (rq_is_sync(rq1) != rq_is_sync(rq2)) |
| 685 | return rq_is_sync(rq1) ? rq1 : rq2; |
| 686 | |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 687 | if ((rq1->cmd_flags ^ rq2->cmd_flags) & REQ_PRIO) |
| 688 | return rq1->cmd_flags & REQ_PRIO ? rq1 : rq2; |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 689 | |
Tejun Heo | 83096eb | 2009-05-07 22:24:39 +0900 | [diff] [blame] | 690 | s1 = blk_rq_pos(rq1); |
| 691 | s2 = blk_rq_pos(rq2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 692 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 693 | /* |
| 694 | * by definition, 1KiB is 2 sectors |
| 695 | */ |
| 696 | back_max = cfqd->cfq_back_max * 2; |
| 697 | |
| 698 | /* |
| 699 | * Strict one way elevator _except_ in the case where we allow |
| 700 | * short backward seeks which are biased as twice the cost of a |
| 701 | * similar forward seek. |
| 702 | */ |
| 703 | if (s1 >= last) |
| 704 | d1 = s1 - last; |
| 705 | else if (s1 + back_max >= last) |
| 706 | d1 = (last - s1) * cfqd->cfq_back_penalty; |
| 707 | else |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 708 | wrap |= CFQ_RQ1_WRAP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 709 | |
| 710 | if (s2 >= last) |
| 711 | d2 = s2 - last; |
| 712 | else if (s2 + back_max >= last) |
| 713 | d2 = (last - s2) * cfqd->cfq_back_penalty; |
| 714 | else |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 715 | wrap |= CFQ_RQ2_WRAP; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | |
| 717 | /* Found required data */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 719 | /* |
| 720 | * By doing switch() on the bit mask "wrap" we avoid having to |
| 721 | * check two variables for all permutations: --> faster! |
| 722 | */ |
| 723 | switch (wrap) { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 724 | case 0: /* common case for CFQ: rq1 and rq2 not wrapped */ |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 725 | if (d1 < d2) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 726 | return rq1; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 727 | else if (d2 < d1) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 728 | return rq2; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 729 | else { |
| 730 | if (s1 >= s2) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 731 | return rq1; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 732 | else |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 733 | return rq2; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 734 | } |
| 735 | |
| 736 | case CFQ_RQ2_WRAP: |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 737 | return rq1; |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 738 | case CFQ_RQ1_WRAP: |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 739 | return rq2; |
| 740 | case (CFQ_RQ1_WRAP|CFQ_RQ2_WRAP): /* both rqs wrapped */ |
Andreas Mohr | e8a9905 | 2006-03-28 08:59:49 +0200 | [diff] [blame] | 741 | default: |
| 742 | /* |
| 743 | * Since both rqs are wrapped, |
| 744 | * start with the one that's further behind head |
| 745 | * (--> only *one* back seek required), |
| 746 | * since back seek takes more time than forward. |
| 747 | */ |
| 748 | if (s1 <= s2) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 749 | return rq1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | else |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 751 | return rq2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 752 | } |
| 753 | } |
| 754 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 755 | /* |
| 756 | * The below is leftmost cache rbtree addon |
| 757 | */ |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 758 | static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root) |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 759 | { |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 760 | /* Service tree is empty */ |
| 761 | if (!root->count) |
| 762 | return NULL; |
| 763 | |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 764 | if (!root->left) |
| 765 | root->left = rb_first(&root->rb); |
| 766 | |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 767 | if (root->left) |
| 768 | return rb_entry(root->left, struct cfq_queue, rb_node); |
| 769 | |
| 770 | return NULL; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 771 | } |
| 772 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 773 | static struct cfq_group *cfq_rb_first_group(struct cfq_rb_root *root) |
| 774 | { |
| 775 | if (!root->left) |
| 776 | root->left = rb_first(&root->rb); |
| 777 | |
| 778 | if (root->left) |
| 779 | return rb_entry_cfqg(root->left); |
| 780 | |
| 781 | return NULL; |
| 782 | } |
| 783 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 784 | static void rb_erase_init(struct rb_node *n, struct rb_root *root) |
| 785 | { |
| 786 | rb_erase(n, root); |
| 787 | RB_CLEAR_NODE(n); |
| 788 | } |
| 789 | |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 790 | static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root) |
| 791 | { |
| 792 | if (root->left == n) |
| 793 | root->left = NULL; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 794 | rb_erase_init(n, &root->rb); |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 795 | --root->count; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 796 | } |
| 797 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | /* |
| 799 | * would be nice to take fifo expire time into account as well |
| 800 | */ |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 801 | static struct request * |
| 802 | cfq_find_next_rq(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 803 | struct request *last) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 804 | { |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 805 | struct rb_node *rbnext = rb_next(&last->rb_node); |
| 806 | struct rb_node *rbprev = rb_prev(&last->rb_node); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 807 | struct request *next = NULL, *prev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 808 | |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 809 | BUG_ON(RB_EMPTY_NODE(&last->rb_node)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 810 | |
| 811 | if (rbprev) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 812 | prev = rb_entry_rq(rbprev); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 813 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 814 | if (rbnext) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 815 | next = rb_entry_rq(rbnext); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 816 | else { |
| 817 | rbnext = rb_first(&cfqq->sort_list); |
| 818 | if (rbnext && rbnext != &last->rb_node) |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 819 | next = rb_entry_rq(rbnext); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 820 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 821 | |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 822 | return cfq_choose_req(cfqd, next, prev, blk_rq_pos(last)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 823 | } |
| 824 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 825 | static unsigned long cfq_slice_offset(struct cfq_data *cfqd, |
| 826 | struct cfq_queue *cfqq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | { |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 828 | /* |
| 829 | * just an approximation, should be ok. |
| 830 | */ |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 831 | return (cfqq->cfqg->nr_cfqq - 1) * (cfq_prio_slice(cfqd, 1, 0) - |
Jens Axboe | 464191c | 2009-11-30 09:38:13 +0100 | [diff] [blame] | 832 | cfq_prio_slice(cfqd, cfq_cfqq_sync(cfqq), cfqq->ioprio)); |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 833 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 834 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 835 | static inline s64 |
| 836 | cfqg_key(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 837 | { |
| 838 | return cfqg->vdisktime - st->min_vdisktime; |
| 839 | } |
| 840 | |
| 841 | static void |
| 842 | __cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 843 | { |
| 844 | struct rb_node **node = &st->rb.rb_node; |
| 845 | struct rb_node *parent = NULL; |
| 846 | struct cfq_group *__cfqg; |
| 847 | s64 key = cfqg_key(st, cfqg); |
| 848 | int left = 1; |
| 849 | |
| 850 | while (*node != NULL) { |
| 851 | parent = *node; |
| 852 | __cfqg = rb_entry_cfqg(parent); |
| 853 | |
| 854 | if (key < cfqg_key(st, __cfqg)) |
| 855 | node = &parent->rb_left; |
| 856 | else { |
| 857 | node = &parent->rb_right; |
| 858 | left = 0; |
| 859 | } |
| 860 | } |
| 861 | |
| 862 | if (left) |
| 863 | st->left = &cfqg->rb_node; |
| 864 | |
| 865 | rb_link_node(&cfqg->rb_node, parent, node); |
| 866 | rb_insert_color(&cfqg->rb_node, &st->rb); |
| 867 | } |
| 868 | |
| 869 | static void |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 870 | cfq_update_group_weight(struct cfq_group *cfqg) |
| 871 | { |
| 872 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
| 873 | if (cfqg->needs_update) { |
| 874 | cfqg->weight = cfqg->new_weight; |
| 875 | cfqg->needs_update = false; |
| 876 | } |
| 877 | } |
| 878 | |
| 879 | static void |
| 880 | cfq_group_service_tree_add(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 881 | { |
| 882 | BUG_ON(!RB_EMPTY_NODE(&cfqg->rb_node)); |
| 883 | |
| 884 | cfq_update_group_weight(cfqg); |
| 885 | __cfq_group_service_tree_add(st, cfqg); |
| 886 | st->total_weight += cfqg->weight; |
| 887 | } |
| 888 | |
| 889 | static void |
| 890 | 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] | 891 | { |
| 892 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 893 | struct cfq_group *__cfqg; |
| 894 | struct rb_node *n; |
| 895 | |
| 896 | cfqg->nr_cfqq++; |
Gui Jianfeng | 760701b | 2010-11-30 20:52:47 +0100 | [diff] [blame] | 897 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 898 | return; |
| 899 | |
| 900 | /* |
| 901 | * Currently put the group at the end. Later implement something |
| 902 | * 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] | 903 | * if group does not loose all if it was not continuously backlogged. |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 904 | */ |
| 905 | n = rb_last(&st->rb); |
| 906 | if (n) { |
| 907 | __cfqg = rb_entry_cfqg(n); |
| 908 | cfqg->vdisktime = __cfqg->vdisktime + CFQ_IDLE_DELAY; |
| 909 | } else |
| 910 | cfqg->vdisktime = st->min_vdisktime; |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 911 | cfq_group_service_tree_add(st, cfqg); |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 912 | } |
| 913 | |
| 914 | static void |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 915 | cfq_group_service_tree_del(struct cfq_rb_root *st, struct cfq_group *cfqg) |
| 916 | { |
| 917 | st->total_weight -= cfqg->weight; |
| 918 | if (!RB_EMPTY_NODE(&cfqg->rb_node)) |
| 919 | cfq_rb_erase(&cfqg->rb_node, st); |
| 920 | } |
| 921 | |
| 922 | static void |
| 923 | 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] | 924 | { |
| 925 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
| 926 | |
| 927 | BUG_ON(cfqg->nr_cfqq < 1); |
| 928 | cfqg->nr_cfqq--; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 929 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 930 | /* If there are other cfq queues under this group, don't delete it */ |
| 931 | if (cfqg->nr_cfqq) |
| 932 | return; |
| 933 | |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 934 | cfq_log_cfqg(cfqd, cfqg, "del_from_rr group"); |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 935 | cfq_group_service_tree_del(st, cfqg); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 936 | cfqg->saved_workload_slice = 0; |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 937 | cfq_blkiocg_update_dequeue_stats(&cfqg->blkg, 1); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 938 | } |
| 939 | |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 940 | static inline unsigned int cfq_cfqq_slice_usage(struct cfq_queue *cfqq, |
| 941 | unsigned int *unaccounted_time) |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 942 | { |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 943 | unsigned int slice_used; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 944 | |
| 945 | /* |
| 946 | * Queue got expired before even a single request completed or |
| 947 | * got expired immediately after first request completion. |
| 948 | */ |
| 949 | if (!cfqq->slice_start || cfqq->slice_start == jiffies) { |
| 950 | /* |
| 951 | * Also charge the seek time incurred to the group, otherwise |
| 952 | * if there are mutiple queues in the group, each can dispatch |
| 953 | * a single request on seeky media and cause lots of seek time |
| 954 | * and group will never know it. |
| 955 | */ |
| 956 | slice_used = max_t(unsigned, (jiffies - cfqq->dispatch_start), |
| 957 | 1); |
| 958 | } else { |
| 959 | slice_used = jiffies - cfqq->slice_start; |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 960 | if (slice_used > cfqq->allocated_slice) { |
| 961 | *unaccounted_time = slice_used - cfqq->allocated_slice; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 962 | slice_used = cfqq->allocated_slice; |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 963 | } |
| 964 | if (time_after(cfqq->slice_start, cfqq->dispatch_start)) |
| 965 | *unaccounted_time += cfqq->slice_start - |
| 966 | cfqq->dispatch_start; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 967 | } |
| 968 | |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 969 | return slice_used; |
| 970 | } |
| 971 | |
| 972 | 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] | 973 | struct cfq_queue *cfqq) |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 974 | { |
| 975 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 976 | unsigned int used_sl, charge, unaccounted_sl = 0; |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 977 | int nr_sync = cfqg->nr_cfqq - cfqg_busy_async_queues(cfqd, cfqg) |
| 978 | - cfqg->service_tree_idle.count; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 979 | |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 980 | BUG_ON(nr_sync < 0); |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 981 | used_sl = charge = cfq_cfqq_slice_usage(cfqq, &unaccounted_sl); |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 982 | |
Vivek Goyal | 02b3508 | 2010-08-23 12:23:53 +0200 | [diff] [blame] | 983 | if (iops_mode(cfqd)) |
| 984 | charge = cfqq->slice_dispatch; |
| 985 | else if (!cfq_cfqq_sync(cfqq) && !nr_sync) |
| 986 | charge = cfqq->allocated_slice; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 987 | |
| 988 | /* Can't update vdisktime while group is on service tree */ |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 989 | cfq_group_service_tree_del(st, cfqg); |
Vivek Goyal | 02b3508 | 2010-08-23 12:23:53 +0200 | [diff] [blame] | 990 | cfqg->vdisktime += cfq_scale_slice(charge, cfqg); |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 991 | /* If a new weight was requested, update now, off tree */ |
| 992 | cfq_group_service_tree_add(st, cfqg); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 993 | |
| 994 | /* This group is being expired. Save the context */ |
| 995 | if (time_after(cfqd->workload_expires, jiffies)) { |
| 996 | cfqg->saved_workload_slice = cfqd->workload_expires |
| 997 | - jiffies; |
| 998 | cfqg->saved_workload = cfqd->serving_type; |
| 999 | cfqg->saved_serving_prio = cfqd->serving_prio; |
| 1000 | } else |
| 1001 | cfqg->saved_workload_slice = 0; |
Vivek Goyal | 2868ef7 | 2009-12-03 12:59:48 -0500 | [diff] [blame] | 1002 | |
| 1003 | cfq_log_cfqg(cfqd, cfqg, "served: vt=%llu min_vt=%llu", cfqg->vdisktime, |
| 1004 | st->min_vdisktime); |
Joe Perches | fd16d26 | 2011-06-13 10:42:49 +0200 | [diff] [blame] | 1005 | cfq_log_cfqq(cfqq->cfqd, cfqq, |
| 1006 | "sl_used=%u disp=%u charge=%u iops=%u sect=%lu", |
| 1007 | used_sl, cfqq->slice_dispatch, charge, |
| 1008 | iops_mode(cfqd), cfqq->nr_sectors); |
Justin TerAvest | 167400d | 2011-03-12 16:54:00 +0100 | [diff] [blame] | 1009 | cfq_blkiocg_update_timeslice_used(&cfqg->blkg, used_sl, |
| 1010 | unaccounted_sl); |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1011 | cfq_blkiocg_set_start_empty_time(&cfqg->blkg); |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1012 | } |
| 1013 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1014 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 1015 | static inline struct cfq_group *cfqg_of_blkg(struct blkio_group *blkg) |
| 1016 | { |
| 1017 | if (blkg) |
| 1018 | return container_of(blkg, struct cfq_group, blkg); |
| 1019 | return NULL; |
| 1020 | } |
| 1021 | |
Paul Bolle | 8aea454 | 2011-06-06 05:07:54 +0200 | [diff] [blame] | 1022 | static void cfq_update_blkio_group_weight(void *key, struct blkio_group *blkg, |
| 1023 | unsigned int weight) |
Vivek Goyal | f8d461d | 2009-12-03 12:59:52 -0500 | [diff] [blame] | 1024 | { |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1025 | struct cfq_group *cfqg = cfqg_of_blkg(blkg); |
| 1026 | cfqg->new_weight = weight; |
| 1027 | cfqg->needs_update = true; |
Vivek Goyal | f8d461d | 2009-12-03 12:59:52 -0500 | [diff] [blame] | 1028 | } |
| 1029 | |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1030 | static void cfq_init_add_cfqg_lists(struct cfq_data *cfqd, |
| 1031 | struct cfq_group *cfqg, struct blkio_cgroup *blkcg) |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1032 | { |
Vivek Goyal | 2208419 | 2009-12-03 12:59:49 -0500 | [diff] [blame] | 1033 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
| 1034 | unsigned int major, minor; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1035 | |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1036 | /* |
| 1037 | * Add group onto cgroup list. It might happen that bdi->dev is |
| 1038 | * not initialized yet. Initialize this new group without major |
| 1039 | * and minor info and this info will be filled in once a new thread |
| 1040 | * comes for IO. |
| 1041 | */ |
| 1042 | if (bdi->dev) { |
Ricky Benitez | a74b2ad | 2010-04-05 18:22:17 +0200 | [diff] [blame] | 1043 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1044 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, |
| 1045 | (void *)cfqd, MKDEV(major, minor)); |
| 1046 | } else |
| 1047 | cfq_blkiocg_add_blkio_group(blkcg, &cfqg->blkg, |
| 1048 | (void *)cfqd, 0); |
| 1049 | |
| 1050 | cfqd->nr_blkcg_linked_grps++; |
| 1051 | cfqg->weight = blkcg_get_weight(blkcg, cfqg->blkg.dev); |
| 1052 | |
| 1053 | /* Add group on cfqd list */ |
| 1054 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); |
| 1055 | } |
| 1056 | |
| 1057 | /* |
| 1058 | * Should be called from sleepable context. No request queue lock as per |
| 1059 | * cpu stats are allocated dynamically and alloc_percpu needs to be called |
| 1060 | * from sleepable context. |
| 1061 | */ |
| 1062 | static struct cfq_group * cfq_alloc_cfqg(struct cfq_data *cfqd) |
| 1063 | { |
| 1064 | struct cfq_group *cfqg = NULL; |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 1065 | int i, j, ret; |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1066 | struct cfq_rb_root *st; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1067 | |
| 1068 | cfqg = kzalloc_node(sizeof(*cfqg), GFP_ATOMIC, cfqd->queue->node); |
| 1069 | if (!cfqg) |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1070 | return NULL; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1071 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1072 | for_each_cfqg_st(cfqg, i, j, st) |
| 1073 | *st = CFQ_RB_ROOT; |
| 1074 | RB_CLEAR_NODE(&cfqg->rb_node); |
| 1075 | |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 1076 | cfqg->ttime.last_end_request = jiffies; |
| 1077 | |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1078 | /* |
| 1079 | * Take the initial reference that will be released on destroy |
| 1080 | * This can be thought of a joint reference by cgroup and |
| 1081 | * elevator which will be dropped by either elevator exit |
| 1082 | * or cgroup deletion path depending on who is exiting first. |
| 1083 | */ |
Shaohua Li | 329a678 | 2011-01-07 08:48:28 +0100 | [diff] [blame] | 1084 | cfqg->ref = 1; |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 1085 | |
| 1086 | ret = blkio_alloc_blkg_stats(&cfqg->blkg); |
| 1087 | if (ret) { |
| 1088 | kfree(cfqg); |
| 1089 | return NULL; |
| 1090 | } |
| 1091 | |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1092 | return cfqg; |
| 1093 | } |
| 1094 | |
| 1095 | static struct cfq_group * |
| 1096 | cfq_find_cfqg(struct cfq_data *cfqd, struct blkio_cgroup *blkcg) |
| 1097 | { |
| 1098 | struct cfq_group *cfqg = NULL; |
| 1099 | void *key = cfqd; |
| 1100 | struct backing_dev_info *bdi = &cfqd->queue->backing_dev_info; |
| 1101 | unsigned int major, minor; |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1102 | |
Vivek Goyal | 180be2a | 2010-09-14 08:47:11 +0200 | [diff] [blame] | 1103 | /* |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1104 | * This is the common case when there are no blkio cgroups. |
| 1105 | * Avoid lookup in this case |
Vivek Goyal | 180be2a | 2010-09-14 08:47:11 +0200 | [diff] [blame] | 1106 | */ |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1107 | if (blkcg == &blkio_root_cgroup) |
| 1108 | cfqg = &cfqd->root_group; |
| 1109 | else |
| 1110 | cfqg = cfqg_of_blkg(blkiocg_lookup_group(blkcg, key)); |
| 1111 | |
| 1112 | if (cfqg && !cfqg->blkg.dev && bdi->dev && dev_name(bdi->dev)) { |
Vivek Goyal | 180be2a | 2010-09-14 08:47:11 +0200 | [diff] [blame] | 1113 | sscanf(dev_name(bdi->dev), "%u:%u", &major, &minor); |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1114 | cfqg->blkg.dev = MKDEV(major, minor); |
| 1115 | } |
Vivek Goyal | 180be2a | 2010-09-14 08:47:11 +0200 | [diff] [blame] | 1116 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1117 | return cfqg; |
| 1118 | } |
| 1119 | |
| 1120 | /* |
Vivek Goyal | 3e59cf9 | 2011-05-19 15:38:21 -0400 | [diff] [blame] | 1121 | * Search for the cfq group current task belongs to. request_queue lock must |
| 1122 | * be held. |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1123 | */ |
Vivek Goyal | 3e59cf9 | 2011-05-19 15:38:21 -0400 | [diff] [blame] | 1124 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd) |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1125 | { |
Vivek Goyal | 70087dc | 2011-05-16 15:24:08 +0200 | [diff] [blame] | 1126 | struct blkio_cgroup *blkcg; |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1127 | struct cfq_group *cfqg = NULL, *__cfqg = NULL; |
| 1128 | struct request_queue *q = cfqd->queue; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1129 | |
| 1130 | rcu_read_lock(); |
Vivek Goyal | 70087dc | 2011-05-16 15:24:08 +0200 | [diff] [blame] | 1131 | blkcg = task_blkio_cgroup(current); |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1132 | cfqg = cfq_find_cfqg(cfqd, blkcg); |
| 1133 | if (cfqg) { |
| 1134 | rcu_read_unlock(); |
| 1135 | return cfqg; |
| 1136 | } |
| 1137 | |
| 1138 | /* |
| 1139 | * Need to allocate a group. Allocation of group also needs allocation |
| 1140 | * of per cpu stats which in-turn takes a mutex() and can block. Hence |
| 1141 | * we need to drop rcu lock and queue_lock before we call alloc. |
| 1142 | * |
| 1143 | * Not taking any queue reference here and assuming that queue is |
| 1144 | * around by the time we return. CFQ queue allocation code does |
| 1145 | * the same. It might be racy though. |
| 1146 | */ |
| 1147 | |
| 1148 | rcu_read_unlock(); |
| 1149 | spin_unlock_irq(q->queue_lock); |
| 1150 | |
| 1151 | cfqg = cfq_alloc_cfqg(cfqd); |
| 1152 | |
| 1153 | spin_lock_irq(q->queue_lock); |
| 1154 | |
| 1155 | rcu_read_lock(); |
| 1156 | blkcg = task_blkio_cgroup(current); |
| 1157 | |
| 1158 | /* |
| 1159 | * If some other thread already allocated the group while we were |
| 1160 | * not holding queue lock, free up the group |
| 1161 | */ |
| 1162 | __cfqg = cfq_find_cfqg(cfqd, blkcg); |
| 1163 | |
| 1164 | if (__cfqg) { |
| 1165 | kfree(cfqg); |
| 1166 | rcu_read_unlock(); |
| 1167 | return __cfqg; |
| 1168 | } |
| 1169 | |
Vivek Goyal | 3e59cf9 | 2011-05-19 15:38:21 -0400 | [diff] [blame] | 1170 | if (!cfqg) |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1171 | cfqg = &cfqd->root_group; |
Vivek Goyal | f469a7b | 2011-05-19 15:38:23 -0400 | [diff] [blame] | 1172 | |
| 1173 | cfq_init_add_cfqg_lists(cfqd, cfqg, blkcg); |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1174 | rcu_read_unlock(); |
| 1175 | return cfqg; |
| 1176 | } |
| 1177 | |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 1178 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) |
| 1179 | { |
Shaohua Li | 329a678 | 2011-01-07 08:48:28 +0100 | [diff] [blame] | 1180 | cfqg->ref++; |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 1181 | return cfqg; |
| 1182 | } |
| 1183 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1184 | static void cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) |
| 1185 | { |
| 1186 | /* Currently, all async queues are mapped to root group */ |
| 1187 | if (!cfq_cfqq_sync(cfqq)) |
| 1188 | cfqg = &cfqq->cfqd->root_group; |
| 1189 | |
| 1190 | cfqq->cfqg = cfqg; |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1191 | /* cfqq reference on cfqg */ |
Shaohua Li | 329a678 | 2011-01-07 08:48:28 +0100 | [diff] [blame] | 1192 | cfqq->cfqg->ref++; |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1193 | } |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1194 | |
| 1195 | static void cfq_put_cfqg(struct cfq_group *cfqg) |
| 1196 | { |
| 1197 | struct cfq_rb_root *st; |
| 1198 | int i, j; |
| 1199 | |
Shaohua Li | 329a678 | 2011-01-07 08:48:28 +0100 | [diff] [blame] | 1200 | BUG_ON(cfqg->ref <= 0); |
| 1201 | cfqg->ref--; |
| 1202 | if (cfqg->ref) |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1203 | return; |
| 1204 | for_each_cfqg_st(cfqg, i, j, st) |
Gui Jianfeng | b54ce60 | 2010-11-30 20:52:46 +0100 | [diff] [blame] | 1205 | BUG_ON(!RB_EMPTY_ROOT(&st->rb)); |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 1206 | free_percpu(cfqg->blkg.stats_cpu); |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1207 | kfree(cfqg); |
| 1208 | } |
| 1209 | |
| 1210 | static void cfq_destroy_cfqg(struct cfq_data *cfqd, struct cfq_group *cfqg) |
| 1211 | { |
| 1212 | /* Something wrong if we are trying to remove same group twice */ |
| 1213 | BUG_ON(hlist_unhashed(&cfqg->cfqd_node)); |
| 1214 | |
| 1215 | hlist_del_init(&cfqg->cfqd_node); |
| 1216 | |
Vivek Goyal | a5395b8 | 2011-08-02 09:24:09 +0200 | [diff] [blame] | 1217 | BUG_ON(cfqd->nr_blkcg_linked_grps <= 0); |
| 1218 | cfqd->nr_blkcg_linked_grps--; |
| 1219 | |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1220 | /* |
| 1221 | * Put the reference taken at the time of creation so that when all |
| 1222 | * queues are gone, group can be destroyed. |
| 1223 | */ |
| 1224 | cfq_put_cfqg(cfqg); |
| 1225 | } |
| 1226 | |
| 1227 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) |
| 1228 | { |
| 1229 | struct hlist_node *pos, *n; |
| 1230 | struct cfq_group *cfqg; |
| 1231 | |
| 1232 | hlist_for_each_entry_safe(cfqg, pos, n, &cfqd->cfqg_list, cfqd_node) { |
| 1233 | /* |
| 1234 | * If cgroup removal path got to blk_group first and removed |
| 1235 | * it from cgroup list, then it will take care of destroying |
| 1236 | * cfqg also. |
| 1237 | */ |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1238 | if (!cfq_blkiocg_del_blkio_group(&cfqg->blkg)) |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1239 | cfq_destroy_cfqg(cfqd, cfqg); |
| 1240 | } |
| 1241 | } |
| 1242 | |
| 1243 | /* |
| 1244 | * Blk cgroup controller notification saying that blkio_group object is being |
| 1245 | * delinked as associated cgroup object is going away. That also means that |
| 1246 | * no new IO will come in this group. So get rid of this group as soon as |
| 1247 | * any pending IO in the group is finished. |
| 1248 | * |
| 1249 | * This function is called under rcu_read_lock(). key is the rcu protected |
| 1250 | * pointer. That means "key" is a valid cfq_data pointer as long as we are rcu |
| 1251 | * read lock. |
| 1252 | * |
| 1253 | * "key" was fetched from blkio_group under blkio_cgroup->lock. That means |
| 1254 | * it should not be NULL as even if elevator was exiting, cgroup deltion |
| 1255 | * path got to it first. |
| 1256 | */ |
Paul Bolle | 8aea454 | 2011-06-06 05:07:54 +0200 | [diff] [blame] | 1257 | static void cfq_unlink_blkio_group(void *key, struct blkio_group *blkg) |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1258 | { |
| 1259 | unsigned long flags; |
| 1260 | struct cfq_data *cfqd = key; |
| 1261 | |
| 1262 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
| 1263 | cfq_destroy_cfqg(cfqd, cfqg_of_blkg(blkg)); |
| 1264 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
| 1265 | } |
| 1266 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1267 | #else /* GROUP_IOSCHED */ |
Vivek Goyal | 3e59cf9 | 2011-05-19 15:38:21 -0400 | [diff] [blame] | 1268 | static struct cfq_group *cfq_get_cfqg(struct cfq_data *cfqd) |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1269 | { |
| 1270 | return &cfqd->root_group; |
| 1271 | } |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 1272 | |
| 1273 | static inline struct cfq_group *cfq_ref_get_cfqg(struct cfq_group *cfqg) |
| 1274 | { |
Dmitry Monakhov | 50eaeb3 | 2010-04-28 19:50:33 +0200 | [diff] [blame] | 1275 | return cfqg; |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 1276 | } |
| 1277 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1278 | static inline void |
| 1279 | cfq_link_cfqq_cfqg(struct cfq_queue *cfqq, struct cfq_group *cfqg) { |
| 1280 | cfqq->cfqg = cfqg; |
| 1281 | } |
| 1282 | |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 1283 | static void cfq_release_cfq_groups(struct cfq_data *cfqd) {} |
| 1284 | static inline void cfq_put_cfqg(struct cfq_group *cfqg) {} |
| 1285 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1286 | #endif /* GROUP_IOSCHED */ |
| 1287 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1288 | /* |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1289 | * The cfqd->service_trees holds all pending cfq_queue's that have |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1290 | * requests waiting to be processed. It is sorted in the order that |
| 1291 | * we will service the queues. |
| 1292 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1293 | 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] | 1294 | bool add_front) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1295 | { |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1296 | struct rb_node **p, *parent; |
| 1297 | struct cfq_queue *__cfqq; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1298 | unsigned long rb_key; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1299 | struct cfq_rb_root *service_tree; |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1300 | int left; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1301 | int new_cfqq = 1; |
Vivek Goyal | ae30c28 | 2009-12-03 12:59:55 -0500 | [diff] [blame] | 1302 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 1303 | service_tree = service_tree_for(cfqq->cfqg, cfqq_prio(cfqq), |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 1304 | cfqq_type(cfqq)); |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1305 | if (cfq_class_idle(cfqq)) { |
| 1306 | rb_key = CFQ_IDLE_DELAY; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1307 | parent = rb_last(&service_tree->rb); |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1308 | if (parent && parent != &cfqq->rb_node) { |
| 1309 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
| 1310 | rb_key += __cfqq->rb_key; |
| 1311 | } else |
| 1312 | rb_key += jiffies; |
| 1313 | } else if (!add_front) { |
Jens Axboe | b9c8946 | 2009-10-06 20:53:44 +0200 | [diff] [blame] | 1314 | /* |
| 1315 | * Get our rb key offset. Subtract any residual slice |
| 1316 | * value carried from last service. A negative resid |
| 1317 | * count indicates slice overrun, and this should position |
| 1318 | * the next service time further away in the tree. |
| 1319 | */ |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1320 | rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies; |
Jens Axboe | b9c8946 | 2009-10-06 20:53:44 +0200 | [diff] [blame] | 1321 | rb_key -= cfqq->slice_resid; |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1322 | cfqq->slice_resid = 0; |
Corrado Zoccolo | 48e025e | 2009-10-05 08:49:23 +0200 | [diff] [blame] | 1323 | } else { |
| 1324 | rb_key = -HZ; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1325 | __cfqq = cfq_rb_first(service_tree); |
Corrado Zoccolo | 48e025e | 2009-10-05 08:49:23 +0200 | [diff] [blame] | 1326 | rb_key += __cfqq ? __cfqq->rb_key : jiffies; |
| 1327 | } |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1328 | |
| 1329 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1330 | new_cfqq = 0; |
Jens Axboe | 99f9628 | 2007-02-05 11:56:25 +0100 | [diff] [blame] | 1331 | /* |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1332 | * same position, nothing more to do |
Jens Axboe | 99f9628 | 2007-02-05 11:56:25 +0100 | [diff] [blame] | 1333 | */ |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1334 | if (rb_key == cfqq->rb_key && |
| 1335 | cfqq->service_tree == service_tree) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1336 | return; |
Jens Axboe | 53b03744 | 2006-07-28 09:48:51 +0200 | [diff] [blame] | 1337 | |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1338 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
| 1339 | cfqq->service_tree = NULL; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1340 | } |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1341 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1342 | left = 1; |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 1343 | parent = NULL; |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1344 | cfqq->service_tree = service_tree; |
| 1345 | p = &service_tree->rb.rb_node; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1346 | while (*p) { |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1347 | struct rb_node **n; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1348 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1349 | parent = *p; |
| 1350 | __cfqq = rb_entry(parent, struct cfq_queue, rb_node); |
| 1351 | |
Jens Axboe | 0c534e0 | 2007-04-18 20:01:57 +0200 | [diff] [blame] | 1352 | /* |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1353 | * sort by key, that represents service time. |
Jens Axboe | 0c534e0 | 2007-04-18 20:01:57 +0200 | [diff] [blame] | 1354 | */ |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1355 | if (time_before(rb_key, __cfqq->rb_key)) |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1356 | n = &(*p)->rb_left; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1357 | else { |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1358 | n = &(*p)->rb_right; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1359 | left = 0; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1360 | } |
Jens Axboe | 67060e3 | 2007-04-18 20:13:32 +0200 | [diff] [blame] | 1361 | |
| 1362 | p = n; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1363 | } |
| 1364 | |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1365 | if (left) |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1366 | service_tree->left = &cfqq->rb_node; |
Jens Axboe | cc09e29 | 2007-04-26 12:53:50 +0200 | [diff] [blame] | 1367 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1368 | cfqq->rb_key = rb_key; |
| 1369 | rb_link_node(&cfqq->rb_node, parent, p); |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1370 | rb_insert_color(&cfqq->rb_node, &service_tree->rb); |
| 1371 | service_tree->count++; |
Namhyung Kim | 20359f2 | 2011-05-24 10:23:22 +0200 | [diff] [blame] | 1372 | if (add_front || !new_cfqq) |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1373 | return; |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1374 | cfq_group_notify_queue_add(cfqd, cfqq->cfqg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1375 | } |
| 1376 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1377 | static struct cfq_queue * |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1378 | cfq_prio_tree_lookup(struct cfq_data *cfqd, struct rb_root *root, |
| 1379 | sector_t sector, struct rb_node **ret_parent, |
| 1380 | struct rb_node ***rb_link) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1381 | { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1382 | struct rb_node **p, *parent; |
| 1383 | struct cfq_queue *cfqq = NULL; |
| 1384 | |
| 1385 | parent = NULL; |
| 1386 | p = &root->rb_node; |
| 1387 | while (*p) { |
| 1388 | struct rb_node **n; |
| 1389 | |
| 1390 | parent = *p; |
| 1391 | cfqq = rb_entry(parent, struct cfq_queue, p_node); |
| 1392 | |
| 1393 | /* |
| 1394 | * Sort strictly based on sector. Smallest to the left, |
| 1395 | * largest to the right. |
| 1396 | */ |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 1397 | if (sector > blk_rq_pos(cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1398 | n = &(*p)->rb_right; |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 1399 | else if (sector < blk_rq_pos(cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1400 | n = &(*p)->rb_left; |
| 1401 | else |
| 1402 | break; |
| 1403 | p = n; |
Jens Axboe | 3ac6c9f | 2009-04-23 12:14:56 +0200 | [diff] [blame] | 1404 | cfqq = NULL; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1405 | } |
| 1406 | |
| 1407 | *ret_parent = parent; |
| 1408 | if (rb_link) |
| 1409 | *rb_link = p; |
Jens Axboe | 3ac6c9f | 2009-04-23 12:14:56 +0200 | [diff] [blame] | 1410 | return cfqq; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1411 | } |
| 1412 | |
| 1413 | static void cfq_prio_tree_add(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 1414 | { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1415 | struct rb_node **p, *parent; |
| 1416 | struct cfq_queue *__cfqq; |
| 1417 | |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1418 | if (cfqq->p_root) { |
| 1419 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 1420 | cfqq->p_root = NULL; |
| 1421 | } |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1422 | |
| 1423 | if (cfq_class_idle(cfqq)) |
| 1424 | return; |
| 1425 | if (!cfqq->next_rq) |
| 1426 | return; |
| 1427 | |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1428 | cfqq->p_root = &cfqd->prio_trees[cfqq->org_ioprio]; |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 1429 | __cfqq = cfq_prio_tree_lookup(cfqd, cfqq->p_root, |
| 1430 | blk_rq_pos(cfqq->next_rq), &parent, &p); |
Jens Axboe | 3ac6c9f | 2009-04-23 12:14:56 +0200 | [diff] [blame] | 1431 | if (!__cfqq) { |
| 1432 | rb_link_node(&cfqq->p_node, parent, p); |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1433 | rb_insert_color(&cfqq->p_node, cfqq->p_root); |
| 1434 | } else |
| 1435 | cfqq->p_root = NULL; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1436 | } |
| 1437 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1438 | /* |
| 1439 | * Update cfqq's position in the service tree. |
| 1440 | */ |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1441 | 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] | 1442 | { |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1443 | /* |
| 1444 | * Resorting requires the cfqq to be on the RR list already. |
| 1445 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1446 | if (cfq_cfqq_on_rr(cfqq)) { |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1447 | cfq_service_tree_add(cfqd, cfqq, 0); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1448 | cfq_prio_tree_add(cfqd, cfqq); |
| 1449 | } |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1450 | } |
| 1451 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1452 | /* |
| 1453 | * 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] | 1454 | * the pending list according to last request service |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1455 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1456 | 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] | 1457 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1458 | cfq_log_cfqq(cfqd, cfqq, "add_to_rr"); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 1459 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
| 1460 | cfq_mark_cfqq_on_rr(cfqq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1461 | cfqd->busy_queues++; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 1462 | if (cfq_cfqq_sync(cfqq)) |
| 1463 | cfqd->busy_sync_queues++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1464 | |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1465 | cfq_resort_rr_list(cfqd, cfqq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1466 | } |
| 1467 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1468 | /* |
| 1469 | * Called when the cfqq no longer has requests pending, remove it from |
| 1470 | * the service tree. |
| 1471 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1472 | 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] | 1473 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1474 | cfq_log_cfqq(cfqd, cfqq, "del_from_rr"); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 1475 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
| 1476 | cfq_clear_cfqq_on_rr(cfqq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 1478 | if (!RB_EMPTY_NODE(&cfqq->rb_node)) { |
| 1479 | cfq_rb_erase(&cfqq->rb_node, cfqq->service_tree); |
| 1480 | cfqq->service_tree = NULL; |
| 1481 | } |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1482 | if (cfqq->p_root) { |
| 1483 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 1484 | cfqq->p_root = NULL; |
| 1485 | } |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1486 | |
Justin TerAvest | 8184f93 | 2011-03-17 16:12:36 +0100 | [diff] [blame] | 1487 | cfq_group_notify_queue_del(cfqd, cfqq->cfqg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1488 | BUG_ON(!cfqd->busy_queues); |
| 1489 | cfqd->busy_queues--; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 1490 | if (cfq_cfqq_sync(cfqq)) |
| 1491 | cfqd->busy_sync_queues--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | } |
| 1493 | |
| 1494 | /* |
| 1495 | * rb tree support functions |
| 1496 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1497 | static void cfq_del_rq_rb(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1499 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1500 | const int sync = rq_is_sync(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1502 | BUG_ON(!cfqq->queued[sync]); |
| 1503 | cfqq->queued[sync]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1505 | elv_rb_del(&cfqq->sort_list, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1506 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1507 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) { |
| 1508 | /* |
| 1509 | * Queue will be deleted from service tree when we actually |
| 1510 | * expire it later. Right now just remove it from prio tree |
| 1511 | * as it is empty. |
| 1512 | */ |
| 1513 | if (cfqq->p_root) { |
| 1514 | rb_erase(&cfqq->p_node, cfqq->p_root); |
| 1515 | cfqq->p_root = NULL; |
| 1516 | } |
| 1517 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1518 | } |
| 1519 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1520 | static void cfq_add_rq_rb(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1521 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1522 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1523 | struct cfq_data *cfqd = cfqq->cfqd; |
Jeff Moyer | 796d511 | 2011-06-02 21:19:05 +0200 | [diff] [blame] | 1524 | struct request *prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1525 | |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 1526 | cfqq->queued[rq_is_sync(rq)]++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | |
Jeff Moyer | 796d511 | 2011-06-02 21:19:05 +0200 | [diff] [blame] | 1528 | elv_rb_add(&cfqq->sort_list, rq); |
Jens Axboe | 5fccbf6 | 2006-10-31 14:21:55 +0100 | [diff] [blame] | 1529 | |
| 1530 | if (!cfq_cfqq_on_rr(cfqq)) |
| 1531 | cfq_add_cfqq_rr(cfqd, cfqq); |
Jens Axboe | 5044eed | 2007-04-25 11:53:48 +0200 | [diff] [blame] | 1532 | |
| 1533 | /* |
| 1534 | * check if this request is a better next-serve candidate |
| 1535 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1536 | prev = cfqq->next_rq; |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 1537 | 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] | 1538 | |
| 1539 | /* |
| 1540 | * adjust priority tree position, if ->next_rq changes |
| 1541 | */ |
| 1542 | if (prev != cfqq->next_rq) |
| 1543 | cfq_prio_tree_add(cfqd, cfqq); |
| 1544 | |
Jens Axboe | 5044eed | 2007-04-25 11:53:48 +0200 | [diff] [blame] | 1545 | BUG_ON(!cfqq->next_rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1546 | } |
| 1547 | |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1548 | 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] | 1549 | { |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 1550 | elv_rb_del(&cfqq->sort_list, rq); |
| 1551 | cfqq->queued[rq_is_sync(rq)]--; |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1552 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
| 1553 | rq_data_dir(rq), rq_is_sync(rq)); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1554 | cfq_add_rq_rb(rq); |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1555 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 1556 | &cfqq->cfqd->serving_group->blkg, rq_data_dir(rq), |
| 1557 | rq_is_sync(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1558 | } |
| 1559 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1560 | static struct request * |
| 1561 | cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1562 | { |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1563 | struct task_struct *tsk = current; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 1564 | struct cfq_io_context *cic; |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1565 | struct cfq_queue *cfqq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1566 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 1567 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 1568 | if (!cic) |
| 1569 | return NULL; |
| 1570 | |
| 1571 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 1572 | if (cfqq) { |
| 1573 | sector_t sector = bio->bi_sector + bio_sectors(bio); |
| 1574 | |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1575 | return elv_rb_find(&cfqq->sort_list, sector); |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 1576 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1578 | return NULL; |
| 1579 | } |
| 1580 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1581 | static void cfq_activate_request(struct request_queue *q, struct request *rq) |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1582 | { |
| 1583 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 1584 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1585 | cfqd->rq_in_driver++; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1586 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "activate rq, drv=%d", |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1587 | cfqd->rq_in_driver); |
Jens Axboe | 25776e3 | 2006-06-01 10:12:26 +0200 | [diff] [blame] | 1588 | |
Tejun Heo | 5b93629 | 2009-05-07 22:24:38 +0900 | [diff] [blame] | 1589 | cfqd->last_position = blk_rq_pos(rq) + blk_rq_sectors(rq); |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1590 | } |
| 1591 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1592 | static void cfq_deactivate_request(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1593 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1594 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1596 | WARN_ON(!cfqd->rq_in_driver); |
| 1597 | cfqd->rq_in_driver--; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1598 | cfq_log_cfqq(cfqd, RQ_CFQQ(rq), "deactivate rq, drv=%d", |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 1599 | cfqd->rq_in_driver); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1600 | } |
| 1601 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1602 | static void cfq_remove_request(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1603 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1604 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1605 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1606 | if (cfqq->next_rq == rq) |
| 1607 | cfqq->next_rq = cfq_find_next_rq(cfqq->cfqd, cfqq, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1608 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1609 | list_del_init(&rq->queuelist); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1610 | cfq_del_rq_rb(rq); |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 1611 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 1612 | cfqq->cfqd->rq_queued--; |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1613 | cfq_blkiocg_update_io_remove_stats(&(RQ_CFQG(rq))->blkg, |
| 1614 | rq_data_dir(rq), rq_is_sync(rq)); |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 1615 | if (rq->cmd_flags & REQ_PRIO) { |
| 1616 | WARN_ON(!cfqq->prio_pending); |
| 1617 | cfqq->prio_pending--; |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 1618 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1619 | } |
| 1620 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1621 | static int cfq_merge(struct request_queue *q, struct request **req, |
| 1622 | struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1623 | { |
| 1624 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 1625 | struct request *__rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1627 | __rq = cfq_find_rq_fmerge(cfqd, bio); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1628 | if (__rq && elv_rq_merge_ok(__rq, bio)) { |
Jens Axboe | 9817064 | 2006-07-28 09:23:08 +0200 | [diff] [blame] | 1629 | *req = __rq; |
| 1630 | return ELEVATOR_FRONT_MERGE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1631 | } |
| 1632 | |
| 1633 | return ELEVATOR_NO_MERGE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1634 | } |
| 1635 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1636 | static void cfq_merged_request(struct request_queue *q, struct request *req, |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1637 | int type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | { |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 1639 | if (type == ELEVATOR_FRONT_MERGE) { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1640 | struct cfq_queue *cfqq = RQ_CFQQ(req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1641 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 1642 | cfq_reposition_rq_rb(cfqq, req); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1643 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1644 | } |
| 1645 | |
Divyesh Shah | 812d402 | 2010-04-08 21:14:23 -0700 | [diff] [blame] | 1646 | static void cfq_bio_merged(struct request_queue *q, struct request *req, |
| 1647 | struct bio *bio) |
| 1648 | { |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1649 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(req))->blkg, |
| 1650 | bio_data_dir(bio), cfq_bio_sync(bio)); |
Divyesh Shah | 812d402 | 2010-04-08 21:14:23 -0700 | [diff] [blame] | 1651 | } |
| 1652 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1653 | static void |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1654 | cfq_merged_requests(struct request_queue *q, struct request *rq, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1655 | struct request *next) |
| 1656 | { |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 1657 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1658 | /* |
| 1659 | * reposition in fifo if next is older than rq |
| 1660 | */ |
| 1661 | if (!list_empty(&rq->queuelist) && !list_empty(&next->queuelist) && |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 1662 | time_before(rq_fifo_time(next), rq_fifo_time(rq))) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1663 | list_move(&rq->queuelist, &next->queuelist); |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 1664 | rq_set_fifo_time(rq, rq_fifo_time(next)); |
| 1665 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1666 | |
Corrado Zoccolo | cf7c25c | 2009-11-08 17:16:46 +0100 | [diff] [blame] | 1667 | if (cfqq->next_rq == next) |
| 1668 | cfqq->next_rq = rq; |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 1669 | cfq_remove_request(next); |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1670 | cfq_blkiocg_update_io_merged_stats(&(RQ_CFQG(rq))->blkg, |
| 1671 | rq_data_dir(next), rq_is_sync(next)); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1672 | } |
| 1673 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 1674 | static int cfq_allow_merge(struct request_queue *q, struct request *rq, |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 1675 | struct bio *bio) |
| 1676 | { |
| 1677 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 1678 | struct cfq_io_context *cic; |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 1679 | struct cfq_queue *cfqq; |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 1680 | |
| 1681 | /* |
Jens Axboe | ec8acb6 | 2007-01-02 18:32:11 +0100 | [diff] [blame] | 1682 | * Disallow merge of a sync bio into an async request. |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 1683 | */ |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 1684 | if (cfq_bio_sync(bio) && !rq_is_sync(rq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 1685 | return false; |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 1686 | |
| 1687 | /* |
Jens Axboe | 719d340 | 2006-12-22 09:38:53 +0100 | [diff] [blame] | 1688 | * Lookup the cfqq that this bio will be queued with. Allow |
| 1689 | * merge only if rq is queued there. |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 1690 | */ |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 1691 | cic = cfq_cic_lookup(cfqd, current->io_context); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 1692 | if (!cic) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 1693 | return false; |
Jens Axboe | 719d340 | 2006-12-22 09:38:53 +0100 | [diff] [blame] | 1694 | |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 1695 | cfqq = cic_to_cfqq(cic, cfq_bio_sync(bio)); |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 1696 | return cfqq == RQ_CFQQ(rq); |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 1697 | } |
| 1698 | |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 1699 | static inline void cfq_del_timer(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 1700 | { |
| 1701 | del_timer(&cfqd->idle_slice_timer); |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 1702 | cfq_blkiocg_update_idle_time_stats(&cfqq->cfqg->blkg); |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 1703 | } |
| 1704 | |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 1705 | static void __cfq_set_active_queue(struct cfq_data *cfqd, |
| 1706 | struct cfq_queue *cfqq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1707 | { |
| 1708 | if (cfqq) { |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 1709 | cfq_log_cfqq(cfqd, cfqq, "set_active wl_prio:%d wl_type:%d", |
| 1710 | cfqd->serving_prio, cfqd->serving_type); |
Justin TerAvest | 62a37f6 | 2011-03-23 08:25:44 +0100 | [diff] [blame] | 1711 | cfq_blkiocg_update_avg_queue_size_stats(&cfqq->cfqg->blkg); |
| 1712 | cfqq->slice_start = 0; |
| 1713 | cfqq->dispatch_start = jiffies; |
| 1714 | cfqq->allocated_slice = 0; |
| 1715 | cfqq->slice_end = 0; |
| 1716 | cfqq->slice_dispatch = 0; |
| 1717 | cfqq->nr_sectors = 0; |
| 1718 | |
| 1719 | cfq_clear_cfqq_wait_request(cfqq); |
| 1720 | cfq_clear_cfqq_must_dispatch(cfqq); |
| 1721 | cfq_clear_cfqq_must_alloc_slice(cfqq); |
| 1722 | cfq_clear_cfqq_fifo_expire(cfqq); |
| 1723 | cfq_mark_cfqq_slice_new(cfqq); |
| 1724 | |
| 1725 | cfq_del_timer(cfqd, cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1726 | } |
| 1727 | |
| 1728 | cfqd->active_queue = cfqq; |
| 1729 | } |
| 1730 | |
| 1731 | /* |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1732 | * current cfqq expired its slice (or was too idle), select new one |
| 1733 | */ |
| 1734 | static void |
| 1735 | __cfq_slice_expired(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 1736 | bool timed_out) |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1737 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1738 | cfq_log_cfqq(cfqd, cfqq, "slice expired t=%d", timed_out); |
| 1739 | |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1740 | if (cfq_cfqq_wait_request(cfqq)) |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 1741 | cfq_del_timer(cfqd, cfqq); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1742 | |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1743 | cfq_clear_cfqq_wait_request(cfqq); |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 1744 | cfq_clear_cfqq_wait_busy(cfqq); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1745 | |
| 1746 | /* |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 1747 | * If this cfqq is shared between multiple processes, check to |
| 1748 | * make sure that those processes are still issuing I/Os within |
| 1749 | * the mean seek distance. If not, it may be time to break the |
| 1750 | * queues apart again. |
| 1751 | */ |
| 1752 | if (cfq_cfqq_coop(cfqq) && CFQQ_SEEKY(cfqq)) |
| 1753 | cfq_mark_cfqq_split_coop(cfqq); |
| 1754 | |
| 1755 | /* |
Jens Axboe | 6084cdd | 2007-04-23 08:25:00 +0200 | [diff] [blame] | 1756 | * 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] | 1757 | */ |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 1758 | if (timed_out) { |
| 1759 | if (cfq_cfqq_slice_new(cfqq)) |
Vivek Goyal | ba5bd52 | 2011-01-19 08:25:02 -0700 | [diff] [blame] | 1760 | cfqq->slice_resid = cfq_scaled_cfqq_slice(cfqd, cfqq); |
Shaohua Li | c553f8e | 2011-01-14 08:41:03 +0100 | [diff] [blame] | 1761 | else |
| 1762 | cfqq->slice_resid = cfqq->slice_end - jiffies; |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 1763 | cfq_log_cfqq(cfqd, cfqq, "resid=%ld", cfqq->slice_resid); |
| 1764 | } |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1765 | |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 1766 | cfq_group_served(cfqd, cfqq->cfqg, cfqq); |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 1767 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1768 | if (cfq_cfqq_on_rr(cfqq) && RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 1769 | cfq_del_cfqq_rr(cfqd, cfqq); |
| 1770 | |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1771 | cfq_resort_rr_list(cfqd, cfqq); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1772 | |
| 1773 | if (cfqq == cfqd->active_queue) |
| 1774 | cfqd->active_queue = NULL; |
| 1775 | |
| 1776 | if (cfqd->active_cic) { |
| 1777 | put_io_context(cfqd->active_cic->ioc); |
| 1778 | cfqd->active_cic = NULL; |
| 1779 | } |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1780 | } |
| 1781 | |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 1782 | 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] | 1783 | { |
| 1784 | struct cfq_queue *cfqq = cfqd->active_queue; |
| 1785 | |
| 1786 | if (cfqq) |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 1787 | __cfq_slice_expired(cfqd, cfqq, timed_out); |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1788 | } |
| 1789 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1790 | /* |
| 1791 | * Get next queue for service. Unless we have a queue preemption, |
| 1792 | * we'll simply select the first cfqq in the service tree. |
| 1793 | */ |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1794 | static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1795 | { |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1796 | struct cfq_rb_root *service_tree = |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 1797 | service_tree_for(cfqd->serving_group, cfqd->serving_prio, |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 1798 | cfqd->serving_type); |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 1799 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1800 | if (!cfqd->rq_queued) |
| 1801 | return NULL; |
| 1802 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 1803 | /* There is nothing to dispatch */ |
| 1804 | if (!service_tree) |
| 1805 | return NULL; |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1806 | if (RB_EMPTY_ROOT(&service_tree->rb)) |
| 1807 | return NULL; |
| 1808 | return cfq_rb_first(service_tree); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1809 | } |
| 1810 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1811 | static struct cfq_queue *cfq_get_next_queue_forced(struct cfq_data *cfqd) |
| 1812 | { |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1813 | struct cfq_group *cfqg; |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1814 | struct cfq_queue *cfqq; |
| 1815 | int i, j; |
| 1816 | struct cfq_rb_root *st; |
| 1817 | |
| 1818 | if (!cfqd->rq_queued) |
| 1819 | return NULL; |
| 1820 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 1821 | cfqg = cfq_get_next_cfqg(cfqd); |
| 1822 | if (!cfqg) |
| 1823 | return NULL; |
| 1824 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1825 | for_each_cfqg_st(cfqg, i, j, st) |
| 1826 | if ((cfqq = cfq_rb_first(st)) != NULL) |
| 1827 | return cfqq; |
| 1828 | return NULL; |
| 1829 | } |
| 1830 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 1831 | /* |
| 1832 | * Get and set a new active queue for service. |
| 1833 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1834 | static struct cfq_queue *cfq_set_active_queue(struct cfq_data *cfqd, |
| 1835 | struct cfq_queue *cfqq) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1836 | { |
Jens Axboe | e00ef79 | 2009-11-04 08:54:55 +0100 | [diff] [blame] | 1837 | if (!cfqq) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1838 | cfqq = cfq_get_next_queue(cfqd); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1839 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1840 | __cfq_set_active_queue(cfqd, cfqq); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 1841 | return cfqq; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1842 | } |
| 1843 | |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1844 | static inline sector_t cfq_dist_from_last(struct cfq_data *cfqd, |
| 1845 | struct request *rq) |
| 1846 | { |
Tejun Heo | 83096eb | 2009-05-07 22:24:39 +0900 | [diff] [blame] | 1847 | if (blk_rq_pos(rq) >= cfqd->last_position) |
| 1848 | return blk_rq_pos(rq) - cfqd->last_position; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1849 | else |
Tejun Heo | 83096eb | 2009-05-07 22:24:39 +0900 | [diff] [blame] | 1850 | return cfqd->last_position - blk_rq_pos(rq); |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1851 | } |
| 1852 | |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 1853 | 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] | 1854 | struct request *rq) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1855 | { |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 1856 | return cfq_dist_from_last(cfqd, rq) <= CFQQ_CLOSE_THR; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1857 | } |
| 1858 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1859 | static struct cfq_queue *cfqq_close(struct cfq_data *cfqd, |
| 1860 | struct cfq_queue *cur_cfqq) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1861 | { |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1862 | struct rb_root *root = &cfqd->prio_trees[cur_cfqq->org_ioprio]; |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1863 | struct rb_node *parent, *node; |
| 1864 | struct cfq_queue *__cfqq; |
| 1865 | sector_t sector = cfqd->last_position; |
| 1866 | |
| 1867 | if (RB_EMPTY_ROOT(root)) |
| 1868 | return NULL; |
| 1869 | |
| 1870 | /* |
| 1871 | * First, if we find a request starting at the end of the last |
| 1872 | * request, choose it. |
| 1873 | */ |
Jens Axboe | f2d1f0a | 2009-04-23 12:19:38 +0200 | [diff] [blame] | 1874 | __cfqq = cfq_prio_tree_lookup(cfqd, root, sector, &parent, NULL); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1875 | if (__cfqq) |
| 1876 | return __cfqq; |
| 1877 | |
| 1878 | /* |
| 1879 | * If the exact sector wasn't found, the parent of the NULL leaf |
| 1880 | * will contain the closest sector. |
| 1881 | */ |
| 1882 | __cfqq = rb_entry(parent, struct cfq_queue, p_node); |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 1883 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1884 | return __cfqq; |
| 1885 | |
Tejun Heo | 2e46e8b | 2009-05-07 22:24:41 +0900 | [diff] [blame] | 1886 | if (blk_rq_pos(__cfqq->next_rq) < sector) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1887 | node = rb_next(&__cfqq->p_node); |
| 1888 | else |
| 1889 | node = rb_prev(&__cfqq->p_node); |
| 1890 | if (!node) |
| 1891 | return NULL; |
| 1892 | |
| 1893 | __cfqq = rb_entry(node, struct cfq_queue, p_node); |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 1894 | if (cfq_rq_close(cfqd, cur_cfqq, __cfqq->next_rq)) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1895 | return __cfqq; |
| 1896 | |
| 1897 | return NULL; |
| 1898 | } |
| 1899 | |
| 1900 | /* |
| 1901 | * cfqd - obvious |
| 1902 | * cur_cfqq - passed in so that we don't decide that the current queue is |
| 1903 | * closely cooperating with itself. |
| 1904 | * |
| 1905 | * So, basically we're assuming that that cur_cfqq has dispatched at least |
| 1906 | * one request, and that cfqd->last_position reflects a position on the disk |
| 1907 | * associated with the I/O issued by cur_cfqq. I'm not sure this is a valid |
| 1908 | * assumption. |
| 1909 | */ |
| 1910 | static struct cfq_queue *cfq_close_cooperator(struct cfq_data *cfqd, |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 1911 | struct cfq_queue *cur_cfqq) |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1912 | { |
| 1913 | struct cfq_queue *cfqq; |
| 1914 | |
Divyesh Shah | 39c01b2 | 2010-03-25 15:45:57 +0100 | [diff] [blame] | 1915 | if (cfq_class_idle(cur_cfqq)) |
| 1916 | return NULL; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 1917 | if (!cfq_cfqq_sync(cur_cfqq)) |
| 1918 | return NULL; |
| 1919 | if (CFQQ_SEEKY(cur_cfqq)) |
| 1920 | return NULL; |
| 1921 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1922 | /* |
Gui Jianfeng | b9d8f4c | 2009-12-08 08:54:17 +0100 | [diff] [blame] | 1923 | * Don't search priority tree if it's the only queue in the group. |
| 1924 | */ |
| 1925 | if (cur_cfqq->cfqg->nr_cfqq == 1) |
| 1926 | return NULL; |
| 1927 | |
| 1928 | /* |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 1929 | * We should notice if some of the queues are cooperating, eg |
| 1930 | * working closely on the same area of the disk. In that case, |
| 1931 | * we can group them together and don't waste time idling. |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1932 | */ |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1933 | cfqq = cfqq_close(cfqd, cur_cfqq); |
| 1934 | if (!cfqq) |
| 1935 | return NULL; |
| 1936 | |
Vivek Goyal | 8682e1f | 2009-12-03 12:59:50 -0500 | [diff] [blame] | 1937 | /* If new queue belongs to different cfq_group, don't choose it */ |
| 1938 | if (cur_cfqq->cfqg != cfqq->cfqg) |
| 1939 | return NULL; |
| 1940 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 1941 | /* |
| 1942 | * It only makes sense to merge sync queues. |
| 1943 | */ |
| 1944 | if (!cfq_cfqq_sync(cfqq)) |
| 1945 | return NULL; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 1946 | if (CFQQ_SEEKY(cfqq)) |
| 1947 | return NULL; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 1948 | |
Corrado Zoccolo | c0324a0 | 2009-10-27 19:16:03 +0100 | [diff] [blame] | 1949 | /* |
| 1950 | * Do not merge queues of different priority classes |
| 1951 | */ |
| 1952 | if (cfq_class_rt(cfqq) != cfq_class_rt(cur_cfqq)) |
| 1953 | return NULL; |
| 1954 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 1955 | return cfqq; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1956 | } |
| 1957 | |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 1958 | /* |
| 1959 | * Determine whether we should enforce idle window for this queue. |
| 1960 | */ |
| 1961 | |
| 1962 | static bool cfq_should_idle(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 1963 | { |
| 1964 | enum wl_prio_t prio = cfqq_prio(cfqq); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 1965 | struct cfq_rb_root *service_tree = cfqq->service_tree; |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 1966 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 1967 | BUG_ON(!service_tree); |
| 1968 | BUG_ON(!service_tree->count); |
| 1969 | |
Vivek Goyal | b6508c1 | 2010-08-23 12:23:33 +0200 | [diff] [blame] | 1970 | if (!cfqd->cfq_slice_idle) |
| 1971 | return false; |
| 1972 | |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 1973 | /* We never do for idle class queues. */ |
| 1974 | if (prio == IDLE_WORKLOAD) |
| 1975 | return false; |
| 1976 | |
| 1977 | /* We do for queues that were marked with idle window flag. */ |
Shaohua Li | 3c764b7 | 2009-12-04 13:12:06 +0100 | [diff] [blame] | 1978 | if (cfq_cfqq_idle_window(cfqq) && |
| 1979 | !(blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag)) |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 1980 | return true; |
| 1981 | |
| 1982 | /* |
| 1983 | * Otherwise, we do only if they are the last ones |
| 1984 | * in their service tree. |
| 1985 | */ |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 1986 | if (service_tree->count == 1 && cfq_cfqq_sync(cfqq) && |
| 1987 | !cfq_io_thinktime_big(cfqd, &service_tree->ttime, false)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 1988 | return true; |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 1989 | cfq_log_cfqq(cfqd, cfqq, "Not idling. st->count:%d", |
| 1990 | service_tree->count); |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 1991 | return false; |
Corrado Zoccolo | a6d44e9 | 2009-10-26 22:45:11 +0100 | [diff] [blame] | 1992 | } |
| 1993 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 1994 | static void cfq_arm_slice_timer(struct cfq_data *cfqd) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 1995 | { |
Jens Axboe | 1792669 | 2007-01-19 11:59:30 +1100 | [diff] [blame] | 1996 | struct cfq_queue *cfqq = cfqd->active_queue; |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 1997 | struct cfq_io_context *cic; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 1998 | unsigned long sl, group_idle = 0; |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 1999 | |
Jens Axboe | a68bbdd | 2008-09-24 13:03:33 +0200 | [diff] [blame] | 2000 | /* |
Jens Axboe | f7d7b7a | 2008-09-25 11:37:50 +0200 | [diff] [blame] | 2001 | * SSD device without seek penalty, disable idling. But only do so |
| 2002 | * for devices that support queuing, otherwise we still have a problem |
| 2003 | * with sync vs async workloads. |
Jens Axboe | a68bbdd | 2008-09-24 13:03:33 +0200 | [diff] [blame] | 2004 | */ |
Jens Axboe | f7d7b7a | 2008-09-25 11:37:50 +0200 | [diff] [blame] | 2005 | if (blk_queue_nonrot(cfqd->queue) && cfqd->hw_tag) |
Jens Axboe | a68bbdd | 2008-09-24 13:03:33 +0200 | [diff] [blame] | 2006 | return; |
| 2007 | |
Jens Axboe | dd67d05 | 2006-06-21 09:36:18 +0200 | [diff] [blame] | 2008 | WARN_ON(!RB_EMPTY_ROOT(&cfqq->sort_list)); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2009 | WARN_ON(cfq_cfqq_slice_new(cfqq)); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2010 | |
| 2011 | /* |
| 2012 | * idle is disabled, either manually or by past process history |
| 2013 | */ |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2014 | if (!cfq_should_idle(cfqd, cfqq)) { |
| 2015 | /* no queue idling. Check for group idling */ |
| 2016 | if (cfqd->cfq_group_idle) |
| 2017 | group_idle = cfqd->cfq_group_idle; |
| 2018 | else |
| 2019 | return; |
| 2020 | } |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2021 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2022 | /* |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 2023 | * still active requests from this queue, don't idle |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2024 | */ |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 2025 | if (cfqq->dispatched) |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2026 | return; |
| 2027 | |
| 2028 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2029 | * task has exited, don't wait |
| 2030 | */ |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 2031 | cic = cfqd->active_cic; |
Nikanth Karthikesan | 66dac98 | 2007-11-27 12:47:04 +0100 | [diff] [blame] | 2032 | if (!cic || !atomic_read(&cic->ioc->nr_tasks)) |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2033 | return; |
| 2034 | |
Corrado Zoccolo | 355b659 | 2009-10-08 08:43:32 +0200 | [diff] [blame] | 2035 | /* |
| 2036 | * If our average think time is larger than the remaining time |
| 2037 | * slice, then don't idle. This avoids overrunning the allotted |
| 2038 | * time slice. |
| 2039 | */ |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 2040 | if (sample_valid(cic->ttime.ttime_samples) && |
| 2041 | (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) { |
Joe Perches | fd16d26 | 2011-06-13 10:42:49 +0200 | [diff] [blame] | 2042 | cfq_log_cfqq(cfqd, cfqq, "Not idling. think_time:%lu", |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 2043 | cic->ttime.ttime_mean); |
Corrado Zoccolo | 355b659 | 2009-10-08 08:43:32 +0200 | [diff] [blame] | 2044 | return; |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 2045 | } |
Corrado Zoccolo | 355b659 | 2009-10-08 08:43:32 +0200 | [diff] [blame] | 2046 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2047 | /* There are other queues in the group, don't do group idle */ |
| 2048 | if (group_idle && cfqq->cfqg->nr_cfqq > 1) |
| 2049 | return; |
| 2050 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2051 | cfq_mark_cfqq_wait_request(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2052 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2053 | if (group_idle) |
| 2054 | sl = cfqd->cfq_group_idle; |
| 2055 | else |
| 2056 | sl = cfqd->cfq_slice_idle; |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 2057 | |
Jens Axboe | 7b14e3b | 2006-02-28 09:35:11 +0100 | [diff] [blame] | 2058 | mod_timer(&cfqd->idle_slice_timer, jiffies + sl); |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 2059 | cfq_blkiocg_update_set_idle_time_stats(&cfqq->cfqg->blkg); |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2060 | cfq_log_cfqq(cfqd, cfqq, "arm_idle: %lu group_idle: %d", sl, |
| 2061 | group_idle ? 1 : 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2062 | } |
| 2063 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2064 | /* |
| 2065 | * Move request from internal lists to the request queue dispatch list. |
| 2066 | */ |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 2067 | static void cfq_dispatch_insert(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2068 | { |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2069 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 2070 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2071 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2072 | cfq_log_cfqq(cfqd, cfqq, "dispatch_insert"); |
| 2073 | |
Jeff Moyer | 06d2188 | 2009-09-11 17:08:59 +0200 | [diff] [blame] | 2074 | cfqq->next_rq = cfq_find_next_rq(cfqd, cfqq, rq); |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 2075 | cfq_remove_request(rq); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2076 | cfqq->dispatched++; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2077 | (RQ_CFQG(rq))->dispatched++; |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 2078 | elv_dispatch_sort(q, rq); |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2079 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 2080 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]++; |
Vivek Goyal | c4e7893 | 2010-08-23 12:25:03 +0200 | [diff] [blame] | 2081 | cfqq->nr_sectors += blk_rq_sectors(rq); |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 2082 | cfq_blkiocg_update_dispatch_stats(&cfqq->cfqg->blkg, blk_rq_bytes(rq), |
Divyesh Shah | 84c124d | 2010-04-09 08:31:19 +0200 | [diff] [blame] | 2083 | rq_data_dir(rq), rq_is_sync(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2084 | } |
| 2085 | |
| 2086 | /* |
| 2087 | * return expired entry, or NULL to just start from scratch in rbtree |
| 2088 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 2089 | static struct request *cfq_check_fifo(struct cfq_queue *cfqq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2090 | { |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 2091 | struct request *rq = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2092 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2093 | if (cfq_cfqq_fifo_expire(cfqq)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2094 | return NULL; |
Jens Axboe | cb88741 | 2007-01-19 12:01:16 +1100 | [diff] [blame] | 2095 | |
| 2096 | cfq_mark_cfqq_fifo_expire(cfqq); |
| 2097 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2098 | if (list_empty(&cfqq->fifo)) |
| 2099 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2100 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2101 | rq = rq_entry_fifo(cfqq->fifo.next); |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 2102 | if (time_before(jiffies, rq_fifo_time(rq))) |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2103 | rq = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2104 | |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 2105 | cfq_log_cfqq(cfqq->cfqd, cfqq, "fifo=%p", rq); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2106 | return rq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2107 | } |
| 2108 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2109 | static inline int |
| 2110 | cfq_prio_to_maxrq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2111 | { |
| 2112 | const int base_rq = cfqd->cfq_slice_async_rq; |
| 2113 | |
| 2114 | WARN_ON(cfqq->ioprio >= IOPRIO_BE_NR); |
| 2115 | |
Namhyung Kim | b9f8ce0 | 2011-05-24 10:23:21 +0200 | [diff] [blame] | 2116 | return 2 * base_rq * (IOPRIO_BE_NR - cfqq->ioprio); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2117 | } |
| 2118 | |
| 2119 | /* |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2120 | * Must be called with the queue_lock held. |
| 2121 | */ |
| 2122 | static int cfqq_process_refs(struct cfq_queue *cfqq) |
| 2123 | { |
| 2124 | int process_refs, io_refs; |
| 2125 | |
| 2126 | io_refs = cfqq->allocated[READ] + cfqq->allocated[WRITE]; |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2127 | process_refs = cfqq->ref - io_refs; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2128 | BUG_ON(process_refs < 0); |
| 2129 | return process_refs; |
| 2130 | } |
| 2131 | |
| 2132 | static void cfq_setup_merge(struct cfq_queue *cfqq, struct cfq_queue *new_cfqq) |
| 2133 | { |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2134 | int process_refs, new_process_refs; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2135 | struct cfq_queue *__cfqq; |
| 2136 | |
Jeff Moyer | c10b61f | 2010-06-17 10:19:11 -0400 | [diff] [blame] | 2137 | /* |
| 2138 | * If there are no process references on the new_cfqq, then it is |
| 2139 | * unsafe to follow the ->new_cfqq chain as other cfqq's in the |
| 2140 | * chain may have dropped their last reference (not just their |
| 2141 | * last process reference). |
| 2142 | */ |
| 2143 | if (!cfqq_process_refs(new_cfqq)) |
| 2144 | return; |
| 2145 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2146 | /* Avoid a circular list and skip interim queue merges */ |
| 2147 | while ((__cfqq = new_cfqq->new_cfqq)) { |
| 2148 | if (__cfqq == cfqq) |
| 2149 | return; |
| 2150 | new_cfqq = __cfqq; |
| 2151 | } |
| 2152 | |
| 2153 | process_refs = cfqq_process_refs(cfqq); |
Jeff Moyer | c10b61f | 2010-06-17 10:19:11 -0400 | [diff] [blame] | 2154 | new_process_refs = cfqq_process_refs(new_cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2155 | /* |
| 2156 | * If the process for the cfqq has gone away, there is no |
| 2157 | * sense in merging the queues. |
| 2158 | */ |
Jeff Moyer | c10b61f | 2010-06-17 10:19:11 -0400 | [diff] [blame] | 2159 | if (process_refs == 0 || new_process_refs == 0) |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2160 | return; |
| 2161 | |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2162 | /* |
| 2163 | * Merge in the direction of the lesser amount of work. |
| 2164 | */ |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2165 | if (new_process_refs >= process_refs) { |
| 2166 | cfqq->new_cfqq = new_cfqq; |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2167 | new_cfqq->ref += process_refs; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2168 | } else { |
| 2169 | new_cfqq->new_cfqq = cfqq; |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2170 | cfqq->ref += new_process_refs; |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 2171 | } |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2172 | } |
| 2173 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2174 | static enum wl_type_t cfq_choose_wl(struct cfq_data *cfqd, |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2175 | struct cfq_group *cfqg, enum wl_prio_t prio) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2176 | { |
| 2177 | struct cfq_queue *queue; |
| 2178 | int i; |
| 2179 | bool key_valid = false; |
| 2180 | unsigned long lowest_key = 0; |
| 2181 | enum wl_type_t cur_best = SYNC_NOIDLE_WORKLOAD; |
| 2182 | |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2183 | for (i = 0; i <= SYNC_WORKLOAD; ++i) { |
| 2184 | /* select the one with lowest rb_key */ |
| 2185 | queue = cfq_rb_first(service_tree_for(cfqg, prio, i)); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2186 | if (queue && |
| 2187 | (!key_valid || time_before(queue->rb_key, lowest_key))) { |
| 2188 | lowest_key = queue->rb_key; |
| 2189 | cur_best = i; |
| 2190 | key_valid = true; |
| 2191 | } |
| 2192 | } |
| 2193 | |
| 2194 | return cur_best; |
| 2195 | } |
| 2196 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2197 | 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] | 2198 | { |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2199 | unsigned slice; |
| 2200 | unsigned count; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2201 | struct cfq_rb_root *st; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2202 | unsigned group_slice; |
Shaohua Li writes | e4ea0c1 | 2010-12-13 14:32:22 +0100 | [diff] [blame] | 2203 | enum wl_prio_t original_prio = cfqd->serving_prio; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2204 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2205 | /* Choose next priority. RT > BE > IDLE */ |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2206 | if (cfq_group_busy_queues_wl(RT_WORKLOAD, cfqd, cfqg)) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2207 | cfqd->serving_prio = RT_WORKLOAD; |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2208 | else if (cfq_group_busy_queues_wl(BE_WORKLOAD, cfqd, cfqg)) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2209 | cfqd->serving_prio = BE_WORKLOAD; |
| 2210 | else { |
| 2211 | cfqd->serving_prio = IDLE_WORKLOAD; |
| 2212 | cfqd->workload_expires = jiffies + 1; |
| 2213 | return; |
| 2214 | } |
| 2215 | |
Shaohua Li writes | e4ea0c1 | 2010-12-13 14:32:22 +0100 | [diff] [blame] | 2216 | if (original_prio != cfqd->serving_prio) |
| 2217 | goto new_workload; |
| 2218 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2219 | /* |
| 2220 | * For RT and BE, we have to choose also the type |
| 2221 | * (SYNC, SYNC_NOIDLE, ASYNC), and to compute a workload |
| 2222 | * expiration time |
| 2223 | */ |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2224 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2225 | count = st->count; |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2226 | |
| 2227 | /* |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2228 | * check workload expiration, and that we still have other queues ready |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2229 | */ |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2230 | if (count && !time_after(jiffies, cfqd->workload_expires)) |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2231 | return; |
| 2232 | |
Shaohua Li writes | e4ea0c1 | 2010-12-13 14:32:22 +0100 | [diff] [blame] | 2233 | new_workload: |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2234 | /* otherwise select new workload type */ |
| 2235 | cfqd->serving_type = |
Vivek Goyal | 65b32a5 | 2009-12-16 17:52:59 -0500 | [diff] [blame] | 2236 | cfq_choose_wl(cfqd, cfqg, cfqd->serving_prio); |
| 2237 | st = service_tree_for(cfqg, cfqd->serving_prio, cfqd->serving_type); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2238 | count = st->count; |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2239 | |
| 2240 | /* |
| 2241 | * the workload slice is computed as a fraction of target latency |
| 2242 | * proportional to the number of queues in that workload, over |
| 2243 | * all the queues in the same priority class |
| 2244 | */ |
Vivek Goyal | 58ff82f | 2009-12-03 12:59:44 -0500 | [diff] [blame] | 2245 | group_slice = cfq_group_slice(cfqd, cfqg); |
| 2246 | |
| 2247 | slice = group_slice * count / |
| 2248 | max_t(unsigned, cfqg->busy_queues_avg[cfqd->serving_prio], |
| 2249 | cfq_group_busy_queues_wl(cfqd->serving_prio, cfqd, cfqg)); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2250 | |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 2251 | if (cfqd->serving_type == ASYNC_WORKLOAD) { |
| 2252 | unsigned int tmp; |
| 2253 | |
| 2254 | /* |
| 2255 | * Async queues are currently system wide. Just taking |
| 2256 | * proportion of queues with-in same group will lead to higher |
| 2257 | * async ratio system wide as generally root group is going |
| 2258 | * to have higher weight. A more accurate thing would be to |
| 2259 | * calculate system wide asnc/sync ratio. |
| 2260 | */ |
| 2261 | tmp = cfq_target_latency * cfqg_busy_async_queues(cfqd, cfqg); |
| 2262 | tmp = tmp/cfqd->busy_queues; |
| 2263 | slice = min_t(unsigned, slice, tmp); |
| 2264 | |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2265 | /* async workload slice is scaled down according to |
| 2266 | * the sync/async slice ratio. */ |
| 2267 | slice = slice * cfqd->cfq_slice[0] / cfqd->cfq_slice[1]; |
Vivek Goyal | f26bd1f | 2009-12-03 12:59:54 -0500 | [diff] [blame] | 2268 | } else |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2269 | /* sync workload slice is at least 2 * cfq_slice_idle */ |
| 2270 | slice = max(slice, 2 * cfqd->cfq_slice_idle); |
| 2271 | |
| 2272 | slice = max_t(unsigned, slice, CFQ_MIN_TT); |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 2273 | cfq_log(cfqd, "workload slice:%d", slice); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2274 | cfqd->workload_expires = jiffies + slice; |
| 2275 | } |
| 2276 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2277 | static struct cfq_group *cfq_get_next_cfqg(struct cfq_data *cfqd) |
| 2278 | { |
| 2279 | struct cfq_rb_root *st = &cfqd->grp_service_tree; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 2280 | struct cfq_group *cfqg; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2281 | |
| 2282 | if (RB_EMPTY_ROOT(&st->rb)) |
| 2283 | return NULL; |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 2284 | cfqg = cfq_rb_first_group(st); |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 2285 | update_min_vdisktime(st); |
| 2286 | return cfqg; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2287 | } |
| 2288 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2289 | static void cfq_choose_cfqg(struct cfq_data *cfqd) |
| 2290 | { |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2291 | struct cfq_group *cfqg = cfq_get_next_cfqg(cfqd); |
| 2292 | |
| 2293 | cfqd->serving_group = cfqg; |
Vivek Goyal | dae739e | 2009-12-03 12:59:45 -0500 | [diff] [blame] | 2294 | |
| 2295 | /* Restore the workload type data */ |
| 2296 | if (cfqg->saved_workload_slice) { |
| 2297 | cfqd->workload_expires = jiffies + cfqg->saved_workload_slice; |
| 2298 | cfqd->serving_type = cfqg->saved_workload; |
| 2299 | cfqd->serving_prio = cfqg->saved_serving_prio; |
Gui Jianfeng | 66ae291 | 2009-12-15 10:08:45 +0100 | [diff] [blame] | 2300 | } else |
| 2301 | cfqd->workload_expires = jiffies - 1; |
| 2302 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 2303 | choose_service_tree(cfqd, cfqg); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2304 | } |
| 2305 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2306 | /* |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2307 | * Select a queue for service. If we have a current active queue, |
| 2308 | * 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] | 2309 | */ |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2310 | static struct cfq_queue *cfq_select_queue(struct cfq_data *cfqd) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2311 | { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2312 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2313 | |
| 2314 | cfqq = cfqd->active_queue; |
| 2315 | if (!cfqq) |
| 2316 | goto new_queue; |
| 2317 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2318 | if (!cfqd->rq_queued) |
| 2319 | return NULL; |
Vivek Goyal | c244bb5 | 2009-12-08 17:52:57 -0500 | [diff] [blame] | 2320 | |
| 2321 | /* |
| 2322 | * We were waiting for group to get backlogged. Expire the queue |
| 2323 | */ |
| 2324 | if (cfq_cfqq_wait_busy(cfqq) && !RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 2325 | goto expire; |
| 2326 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2327 | /* |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2328 | * 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] | 2329 | */ |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 2330 | if (cfq_slice_used(cfqq) && !cfq_cfqq_must_dispatch(cfqq)) { |
| 2331 | /* |
| 2332 | * If slice had not expired at the completion of last request |
| 2333 | * we might not have turned on wait_busy flag. Don't expire |
| 2334 | * the queue yet. Allow the group to get backlogged. |
| 2335 | * |
| 2336 | * The very fact that we have used the slice, that means we |
| 2337 | * have been idling all along on this queue and it should be |
| 2338 | * ok to wait for this request to complete. |
| 2339 | */ |
Vivek Goyal | 82bbbf2 | 2009-12-10 19:25:41 +0100 | [diff] [blame] | 2340 | if (cfqq->cfqg->nr_cfqq == 1 && RB_EMPTY_ROOT(&cfqq->sort_list) |
| 2341 | && cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
| 2342 | cfqq = NULL; |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 2343 | goto keep_queue; |
Vivek Goyal | 82bbbf2 | 2009-12-10 19:25:41 +0100 | [diff] [blame] | 2344 | } else |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2345 | goto check_group_idle; |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 2346 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2347 | |
| 2348 | /* |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2349 | * The active queue has requests and isn't expired, allow it to |
| 2350 | * dispatch. |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2351 | */ |
Jens Axboe | dd67d05 | 2006-06-21 09:36:18 +0200 | [diff] [blame] | 2352 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2353 | goto keep_queue; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2354 | |
| 2355 | /* |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2356 | * If another queue has a request waiting within our mean seek |
| 2357 | * distance, let it run. The expire code will check for close |
| 2358 | * cooperators and put the close queue at the front of the service |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2359 | * tree. If possible, merge the expiring queue with the new cfqq. |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2360 | */ |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 2361 | new_cfqq = cfq_close_cooperator(cfqd, cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2362 | if (new_cfqq) { |
| 2363 | if (!cfqq->new_cfqq) |
| 2364 | cfq_setup_merge(cfqq, new_cfqq); |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2365 | goto expire; |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2366 | } |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2367 | |
| 2368 | /* |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 2369 | * No requests pending. If the active queue still has requests in |
| 2370 | * flight or is idling for a new request, allow either of these |
| 2371 | * conditions to happen (or time out) before selecting a new queue. |
| 2372 | */ |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2373 | if (timer_pending(&cfqd->idle_slice_timer)) { |
| 2374 | cfqq = NULL; |
| 2375 | goto keep_queue; |
| 2376 | } |
| 2377 | |
Shaohua Li | 8e1ac66 | 2010-11-08 15:01:04 +0100 | [diff] [blame] | 2378 | /* |
| 2379 | * This is a deep seek queue, but the device is much faster than |
| 2380 | * the queue can deliver, don't idle |
| 2381 | **/ |
| 2382 | if (CFQQ_SEEKY(cfqq) && cfq_cfqq_idle_window(cfqq) && |
| 2383 | (cfq_cfqq_slice_new(cfqq) || |
| 2384 | (cfqq->slice_end - jiffies > jiffies - cfqq->slice_start))) { |
| 2385 | cfq_clear_cfqq_deep(cfqq); |
| 2386 | cfq_clear_cfqq_idle_window(cfqq); |
| 2387 | } |
| 2388 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 2389 | if (cfqq->dispatched && cfq_should_idle(cfqd, cfqq)) { |
| 2390 | cfqq = NULL; |
| 2391 | goto keep_queue; |
| 2392 | } |
| 2393 | |
| 2394 | /* |
| 2395 | * If group idle is enabled and there are requests dispatched from |
| 2396 | * this group, wait for requests to complete. |
| 2397 | */ |
| 2398 | check_group_idle: |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 2399 | if (cfqd->cfq_group_idle && cfqq->cfqg->nr_cfqq == 1 && |
| 2400 | cfqq->cfqg->dispatched && |
| 2401 | !cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) { |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2402 | cfqq = NULL; |
| 2403 | goto keep_queue; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2404 | } |
| 2405 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2406 | expire: |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2407 | cfq_slice_expired(cfqd, 0); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2408 | new_queue: |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2409 | /* |
| 2410 | * Current queue expired. Check if we have to switch to a new |
| 2411 | * service tree |
| 2412 | */ |
| 2413 | if (!new_cfqq) |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2414 | cfq_choose_cfqg(cfqd); |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 2415 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 2416 | cfqq = cfq_set_active_queue(cfqd, new_cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2417 | keep_queue: |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2418 | return cfqq; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2419 | } |
| 2420 | |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 2421 | static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq) |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2422 | { |
| 2423 | int dispatched = 0; |
| 2424 | |
| 2425 | while (cfqq->next_rq) { |
| 2426 | cfq_dispatch_insert(cfqq->cfqd->queue, cfqq->next_rq); |
| 2427 | dispatched++; |
| 2428 | } |
| 2429 | |
| 2430 | BUG_ON(!list_empty(&cfqq->fifo)); |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2431 | |
| 2432 | /* 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] | 2433 | __cfq_slice_expired(cfqq->cfqd, cfqq, 0); |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2434 | return dispatched; |
| 2435 | } |
| 2436 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2437 | /* |
| 2438 | * Drain our current requests. Used for barriers and when switching |
| 2439 | * io schedulers on-the-fly. |
| 2440 | */ |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2441 | static int cfq_forced_dispatch(struct cfq_data *cfqd) |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2442 | { |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 2443 | struct cfq_queue *cfqq; |
Jens Axboe | d9e7620 | 2007-04-20 14:27:50 +0200 | [diff] [blame] | 2444 | int dispatched = 0; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 2445 | |
Divyesh Shah | 3440c49 | 2010-04-09 09:29:57 +0200 | [diff] [blame] | 2446 | /* Expire the timeslice of the current active queue first */ |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2447 | cfq_slice_expired(cfqd, 0); |
Divyesh Shah | 3440c49 | 2010-04-09 09:29:57 +0200 | [diff] [blame] | 2448 | while ((cfqq = cfq_get_next_queue_forced(cfqd)) != NULL) { |
| 2449 | __cfq_set_active_queue(cfqd, cfqq); |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2450 | dispatched += __cfq_forced_dispatch_cfqq(cfqq); |
Divyesh Shah | 3440c49 | 2010-04-09 09:29:57 +0200 | [diff] [blame] | 2451 | } |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2452 | |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2453 | BUG_ON(cfqd->busy_queues); |
| 2454 | |
Jeff Moyer | 6923715 | 2009-06-12 15:29:30 +0200 | [diff] [blame] | 2455 | cfq_log(cfqd, "forced_dispatch=%d", dispatched); |
Tejun Heo | 1b5ed5e1 | 2005-11-10 08:49:19 +0100 | [diff] [blame] | 2456 | return dispatched; |
| 2457 | } |
| 2458 | |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2459 | static inline bool cfq_slice_used_soon(struct cfq_data *cfqd, |
| 2460 | struct cfq_queue *cfqq) |
| 2461 | { |
| 2462 | /* the queue hasn't finished any request, can't estimate */ |
| 2463 | if (cfq_cfqq_slice_new(cfqq)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2464 | return true; |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2465 | if (time_after(jiffies + cfqd->cfq_slice_idle * cfqq->dispatched, |
| 2466 | cfqq->slice_end)) |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2467 | return true; |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2468 | |
Shaohua Li | c1e4475 | 2010-11-08 15:01:02 +0100 | [diff] [blame] | 2469 | return false; |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2470 | } |
| 2471 | |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2472 | 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] | 2473 | { |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2474 | unsigned int max_dispatch; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2475 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2476 | /* |
Jens Axboe | 5ad531d | 2009-07-03 12:57:48 +0200 | [diff] [blame] | 2477 | * Drain async requests before we start sync IO |
| 2478 | */ |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 2479 | 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] | 2480 | return false; |
Jens Axboe | 5ad531d | 2009-07-03 12:57:48 +0200 | [diff] [blame] | 2481 | |
| 2482 | /* |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2483 | * If this is an async queue and we have sync IO in flight, let it wait |
| 2484 | */ |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 2485 | if (cfqd->rq_in_flight[BLK_RW_SYNC] && !cfq_cfqq_sync(cfqq)) |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2486 | return false; |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2487 | |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2488 | max_dispatch = max_t(unsigned int, cfqd->cfq_quantum / 2, 1); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2489 | if (cfq_class_idle(cfqq)) |
| 2490 | max_dispatch = 1; |
| 2491 | |
| 2492 | /* |
| 2493 | * Does this cfqq already have too much IO in flight? |
| 2494 | */ |
| 2495 | if (cfqq->dispatched >= max_dispatch) { |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2496 | bool promote_sync = false; |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2497 | /* |
| 2498 | * idle queue must always only have a single IO in flight |
| 2499 | */ |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2500 | if (cfq_class_idle(cfqq)) |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2501 | return false; |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 2502 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2503 | /* |
Li, Shaohua | c4ade94 | 2011-03-23 08:30:34 +0100 | [diff] [blame] | 2504 | * If there is only one sync queue |
| 2505 | * we can ignore async queue here and give the sync |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2506 | * queue no dispatch limit. The reason is a sync queue can |
| 2507 | * preempt async queue, limiting the sync queue doesn't make |
| 2508 | * sense. This is useful for aiostress test. |
| 2509 | */ |
Li, Shaohua | c4ade94 | 2011-03-23 08:30:34 +0100 | [diff] [blame] | 2510 | if (cfq_cfqq_sync(cfqq) && cfqd->busy_sync_queues == 1) |
| 2511 | promote_sync = true; |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2512 | |
| 2513 | /* |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2514 | * We have other queues, don't allow more IO from this one |
| 2515 | */ |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2516 | if (cfqd->busy_queues > 1 && cfq_slice_used_soon(cfqd, cfqq) && |
| 2517 | !promote_sync) |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2518 | return false; |
Jens Axboe | 9ede209 | 2007-01-19 12:11:44 +1100 | [diff] [blame] | 2519 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2520 | /* |
Shaohua Li | 474b18c | 2009-12-03 12:58:05 +0100 | [diff] [blame] | 2521 | * Sole queue user, no limit |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 2522 | */ |
Shaohua Li | ef8a41d | 2011-03-07 09:26:29 +0100 | [diff] [blame] | 2523 | if (cfqd->busy_queues == 1 || promote_sync) |
Shaohua Li | abc3c74 | 2010-03-01 09:20:54 +0100 | [diff] [blame] | 2524 | max_dispatch = -1; |
| 2525 | else |
| 2526 | /* |
| 2527 | * Normally we start throttling cfqq when cfq_quantum/2 |
| 2528 | * requests have been dispatched. But we can drive |
| 2529 | * deeper queue depths at the beginning of slice |
| 2530 | * subjected to upper limit of cfq_quantum. |
| 2531 | * */ |
| 2532 | max_dispatch = cfqd->cfq_quantum; |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2533 | } |
| 2534 | |
| 2535 | /* |
| 2536 | * Async queues must wait a bit before being allowed dispatch. |
| 2537 | * We also ramp up the dispatch depth gradually for async IO, |
| 2538 | * based on the last sync IO we serviced |
| 2539 | */ |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 2540 | if (!cfq_cfqq_sync(cfqq) && cfqd->cfq_latency) { |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 2541 | unsigned long last_sync = jiffies - cfqd->last_delayed_sync; |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2542 | unsigned int depth; |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 2543 | |
Jens Axboe | 61f0c1d | 2009-10-03 19:46:03 +0200 | [diff] [blame] | 2544 | depth = last_sync / cfqd->cfq_slice[1]; |
Jens Axboe | e00c54c | 2009-10-04 20:36:19 +0200 | [diff] [blame] | 2545 | if (!depth && !cfqq->dispatched) |
| 2546 | depth = 1; |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2547 | if (depth < max_dispatch) |
| 2548 | max_dispatch = depth; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2549 | } |
| 2550 | |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2551 | /* |
| 2552 | * If we're below the current max, allow a dispatch |
| 2553 | */ |
| 2554 | return cfqq->dispatched < max_dispatch; |
| 2555 | } |
| 2556 | |
| 2557 | /* |
| 2558 | * Dispatch a request from cfqq, moving them to the request queue |
| 2559 | * dispatch list. |
| 2560 | */ |
| 2561 | static bool cfq_dispatch_request(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2562 | { |
| 2563 | struct request *rq; |
| 2564 | |
| 2565 | BUG_ON(RB_EMPTY_ROOT(&cfqq->sort_list)); |
| 2566 | |
| 2567 | if (!cfq_may_dispatch(cfqd, cfqq)) |
| 2568 | return false; |
| 2569 | |
| 2570 | /* |
| 2571 | * follow expired path, else get first next available |
| 2572 | */ |
| 2573 | rq = cfq_check_fifo(cfqq); |
| 2574 | if (!rq) |
| 2575 | rq = cfqq->next_rq; |
| 2576 | |
| 2577 | /* |
| 2578 | * insert request into driver dispatch list |
| 2579 | */ |
| 2580 | cfq_dispatch_insert(cfqd->queue, rq); |
| 2581 | |
| 2582 | if (!cfqd->active_cic) { |
| 2583 | struct cfq_io_context *cic = RQ_CIC(rq); |
| 2584 | |
| 2585 | atomic_long_inc(&cic->ioc->refcount); |
| 2586 | cfqd->active_cic = cic; |
| 2587 | } |
| 2588 | |
| 2589 | return true; |
| 2590 | } |
| 2591 | |
| 2592 | /* |
| 2593 | * Find the cfqq that we need to service and move a request from that to the |
| 2594 | * dispatch list |
| 2595 | */ |
| 2596 | static int cfq_dispatch_requests(struct request_queue *q, int force) |
| 2597 | { |
| 2598 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 2599 | struct cfq_queue *cfqq; |
| 2600 | |
| 2601 | if (!cfqd->busy_queues) |
| 2602 | return 0; |
| 2603 | |
| 2604 | if (unlikely(force)) |
| 2605 | return cfq_forced_dispatch(cfqd); |
| 2606 | |
| 2607 | cfqq = cfq_select_queue(cfqd); |
| 2608 | if (!cfqq) |
Jens Axboe | 8e29675 | 2009-10-03 16:26:03 +0200 | [diff] [blame] | 2609 | return 0; |
| 2610 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2611 | /* |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2612 | * Dispatch a request from this cfqq, if it is allowed |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2613 | */ |
Jens Axboe | 0b182d6 | 2009-10-06 20:49:37 +0200 | [diff] [blame] | 2614 | if (!cfq_dispatch_request(cfqd, cfqq)) |
| 2615 | return 0; |
| 2616 | |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2617 | cfqq->slice_dispatch++; |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 2618 | cfq_clear_cfqq_must_dispatch(cfqq); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2619 | |
| 2620 | /* |
| 2621 | * expire an async queue immediately if it has used up its slice. idle |
| 2622 | * queue always expire after 1 dispatch round. |
| 2623 | */ |
| 2624 | if (cfqd->busy_queues > 1 && ((!cfq_cfqq_sync(cfqq) && |
| 2625 | cfqq->slice_dispatch >= cfq_prio_to_maxrq(cfqd, cfqq)) || |
| 2626 | cfq_class_idle(cfqq))) { |
| 2627 | cfqq->slice_end = jiffies + 1; |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2628 | cfq_slice_expired(cfqd, 0); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2629 | } |
| 2630 | |
Shan Wei | b217a90 | 2009-09-01 10:06:42 +0200 | [diff] [blame] | 2631 | cfq_log_cfqq(cfqd, cfqq, "dispatched a request"); |
Jens Axboe | 2f5cb73 | 2009-04-07 08:51:19 +0200 | [diff] [blame] | 2632 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2633 | } |
| 2634 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2635 | /* |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 2636 | * task holds one reference to the queue, dropped when task exits. each rq |
| 2637 | * 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] | 2638 | * |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 2639 | * 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] | 2640 | * queue lock must be held here. |
| 2641 | */ |
| 2642 | static void cfq_put_queue(struct cfq_queue *cfqq) |
| 2643 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2644 | struct cfq_data *cfqd = cfqq->cfqd; |
Justin TerAvest | 0bbfeb8 | 2011-03-01 15:05:08 -0500 | [diff] [blame] | 2645 | struct cfq_group *cfqg; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2646 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2647 | BUG_ON(cfqq->ref <= 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2648 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2649 | cfqq->ref--; |
| 2650 | if (cfqq->ref) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2651 | return; |
| 2652 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 2653 | cfq_log_cfqq(cfqd, cfqq, "put_queue"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2654 | BUG_ON(rb_first(&cfqq->sort_list)); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2655 | BUG_ON(cfqq->allocated[READ] + cfqq->allocated[WRITE]); |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 2656 | cfqg = cfqq->cfqg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2657 | |
Jens Axboe | 28f95cbc | 2007-01-19 12:09:53 +1100 | [diff] [blame] | 2658 | if (unlikely(cfqd->active_queue == cfqq)) { |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 2659 | __cfq_slice_expired(cfqd, cfqq, 0); |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 2660 | cfq_schedule_dispatch(cfqd); |
Jens Axboe | 28f95cbc | 2007-01-19 12:09:53 +1100 | [diff] [blame] | 2661 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2662 | |
Vivek Goyal | f04a642 | 2009-12-03 12:59:40 -0500 | [diff] [blame] | 2663 | BUG_ON(cfq_cfqq_on_rr(cfqq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2664 | kmem_cache_free(cfq_pool, cfqq); |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 2665 | cfq_put_cfqg(cfqg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2666 | } |
| 2667 | |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 2668 | /* |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2669 | * Call func for each cic attached to this ioc. |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2670 | */ |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2671 | static void |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2672 | call_for_each_cic(struct io_context *ioc, |
| 2673 | void (*func)(struct io_context *, struct cfq_io_context *)) |
| 2674 | { |
Jens Axboe | 5f45c69 | 2011-04-19 09:10:35 +0200 | [diff] [blame] | 2675 | struct cfq_io_context *cic; |
| 2676 | struct hlist_node *n; |
| 2677 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2678 | rcu_read_lock(); |
Jens Axboe | 5f45c69 | 2011-04-19 09:10:35 +0200 | [diff] [blame] | 2679 | |
| 2680 | hlist_for_each_entry_rcu(cic, n, &ioc->cic_list, cic_list) |
| 2681 | func(ioc, cic); |
| 2682 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2683 | rcu_read_unlock(); |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2684 | } |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2685 | |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2686 | static void cfq_cic_free_rcu(struct rcu_head *head) |
| 2687 | { |
| 2688 | struct cfq_io_context *cic; |
| 2689 | |
| 2690 | cic = container_of(head, struct cfq_io_context, rcu_head); |
| 2691 | |
| 2692 | kmem_cache_free(cfq_ioc_pool, cic); |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 2693 | elv_ioc_count_dec(cfq_ioc_count); |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2694 | |
Jens Axboe | 9a11b4e | 2008-05-29 09:32:08 +0200 | [diff] [blame] | 2695 | if (ioc_gone) { |
| 2696 | /* |
| 2697 | * CFQ scheduler is exiting, grab exit lock and check |
| 2698 | * the pending io context count. If it hits zero, |
| 2699 | * complete ioc_gone and set it back to NULL |
| 2700 | */ |
| 2701 | spin_lock(&ioc_gone_lock); |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 2702 | if (ioc_gone && !elv_ioc_count_read(cfq_ioc_count)) { |
Jens Axboe | 9a11b4e | 2008-05-29 09:32:08 +0200 | [diff] [blame] | 2703 | complete(ioc_gone); |
| 2704 | ioc_gone = NULL; |
| 2705 | } |
| 2706 | spin_unlock(&ioc_gone_lock); |
| 2707 | } |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2708 | } |
| 2709 | |
| 2710 | static void cfq_cic_free(struct cfq_io_context *cic) |
| 2711 | { |
| 2712 | call_rcu(&cic->rcu_head, cfq_cic_free_rcu); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2713 | } |
| 2714 | |
| 2715 | static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic) |
| 2716 | { |
| 2717 | unsigned long flags; |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2718 | unsigned long dead_key = (unsigned long) cic->key; |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2719 | |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2720 | BUG_ON(!(dead_key & CIC_DEAD_KEY)); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2721 | |
| 2722 | spin_lock_irqsave(&ioc->lock, flags); |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 2723 | radix_tree_delete(&ioc->radix_root, dead_key >> CIC_DEAD_INDEX_SHIFT); |
Jens Axboe | ffc4e75 | 2008-02-19 10:02:29 +0100 | [diff] [blame] | 2724 | hlist_del_rcu(&cic->cic_list); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2725 | spin_unlock_irqrestore(&ioc->lock, flags); |
| 2726 | |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2727 | cfq_cic_free(cic); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2728 | } |
| 2729 | |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 2730 | /* |
| 2731 | * Must be called with rcu_read_lock() held or preemption otherwise disabled. |
| 2732 | * Only two callers of this - ->dtor() which is called with the rcu_read_lock(), |
| 2733 | * and ->trim() which is called with the task lock held |
| 2734 | */ |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 2735 | static void cfq_free_io_context(struct io_context *ioc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2736 | { |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2737 | /* |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 2738 | * ioc->refcount is zero here, or we are called from elv_unregister(), |
| 2739 | * so no more cic's are allowed to be linked into this ioc. So it |
| 2740 | * should be ok to iterate over the known list, we will see all cic's |
| 2741 | * since no new ones are added. |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2742 | */ |
Jens Axboe | 5f45c69 | 2011-04-19 09:10:35 +0200 | [diff] [blame] | 2743 | call_for_each_cic(ioc, cic_free_func); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2744 | } |
| 2745 | |
Shaohua Li | d02a2c0 | 2010-05-25 10:16:53 +0200 | [diff] [blame] | 2746 | static void cfq_put_cooperator(struct cfq_queue *cfqq) |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2747 | { |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2748 | struct cfq_queue *__cfqq, *next; |
| 2749 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2750 | /* |
| 2751 | * If this queue was scheduled to merge with another queue, be |
| 2752 | * sure to drop the reference taken on that queue (and others in |
| 2753 | * the merge chain). See cfq_setup_merge and cfq_merge_cfqqs. |
| 2754 | */ |
| 2755 | __cfqq = cfqq->new_cfqq; |
| 2756 | while (__cfqq) { |
| 2757 | if (__cfqq == cfqq) { |
| 2758 | WARN(1, "cfqq->new_cfqq loop detected\n"); |
| 2759 | break; |
| 2760 | } |
| 2761 | next = __cfqq->new_cfqq; |
| 2762 | cfq_put_queue(__cfqq); |
| 2763 | __cfqq = next; |
| 2764 | } |
Shaohua Li | d02a2c0 | 2010-05-25 10:16:53 +0200 | [diff] [blame] | 2765 | } |
| 2766 | |
| 2767 | static void cfq_exit_cfqq(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 2768 | { |
| 2769 | if (unlikely(cfqq == cfqd->active_queue)) { |
| 2770 | __cfq_slice_expired(cfqd, cfqq, 0); |
| 2771 | cfq_schedule_dispatch(cfqd); |
| 2772 | } |
| 2773 | |
| 2774 | cfq_put_cooperator(cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 2775 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2776 | cfq_put_queue(cfqq); |
| 2777 | } |
| 2778 | |
| 2779 | static void __cfq_exit_single_io_context(struct cfq_data *cfqd, |
| 2780 | struct cfq_io_context *cic) |
| 2781 | { |
Fabio Checconi | 4faa3c8 | 2008-04-10 08:28:01 +0200 | [diff] [blame] | 2782 | struct io_context *ioc = cic->ioc; |
| 2783 | |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 2784 | list_del_init(&cic->queue_list); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2785 | |
| 2786 | /* |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2787 | * Make sure dead mark is seen for dead queues |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2788 | */ |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 2789 | smp_wmb(); |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2790 | cic->key = cfqd_dead_key(cfqd); |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 2791 | |
Shaohua Li | 3181faa | 2011-06-27 09:03:47 +0200 | [diff] [blame] | 2792 | rcu_read_lock(); |
Jens Axboe | 9b50902d | 2011-06-05 06:01:13 +0200 | [diff] [blame] | 2793 | if (rcu_dereference(ioc->ioc_data) == cic) { |
Shaohua Li | 3181faa | 2011-06-27 09:03:47 +0200 | [diff] [blame] | 2794 | rcu_read_unlock(); |
Jens Axboe | 9b50902d | 2011-06-05 06:01:13 +0200 | [diff] [blame] | 2795 | spin_lock(&ioc->lock); |
Fabio Checconi | 4faa3c8 | 2008-04-10 08:28:01 +0200 | [diff] [blame] | 2796 | rcu_assign_pointer(ioc->ioc_data, NULL); |
Jens Axboe | 9b50902d | 2011-06-05 06:01:13 +0200 | [diff] [blame] | 2797 | spin_unlock(&ioc->lock); |
Shaohua Li | 3181faa | 2011-06-27 09:03:47 +0200 | [diff] [blame] | 2798 | } else |
| 2799 | rcu_read_unlock(); |
Fabio Checconi | 4faa3c8 | 2008-04-10 08:28:01 +0200 | [diff] [blame] | 2800 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 2801 | if (cic->cfqq[BLK_RW_ASYNC]) { |
| 2802 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_ASYNC]); |
| 2803 | cic->cfqq[BLK_RW_ASYNC] = NULL; |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2804 | } |
| 2805 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 2806 | if (cic->cfqq[BLK_RW_SYNC]) { |
| 2807 | cfq_exit_cfqq(cfqd, cic->cfqq[BLK_RW_SYNC]); |
| 2808 | cic->cfqq[BLK_RW_SYNC] = NULL; |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2809 | } |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2810 | } |
| 2811 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2812 | static void cfq_exit_single_io_context(struct io_context *ioc, |
| 2813 | struct cfq_io_context *cic) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2814 | { |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2815 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2816 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 2817 | if (cfqd) { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 2818 | struct request_queue *q = cfqd->queue; |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2819 | unsigned long flags; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2820 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2821 | spin_lock_irqsave(q->queue_lock, flags); |
Jens Axboe | 62c1fe9 | 2008-12-15 21:19:25 +0100 | [diff] [blame] | 2822 | |
| 2823 | /* |
| 2824 | * Ensure we get a fresh copy of the ->key to prevent |
| 2825 | * race between exiting task and queue |
| 2826 | */ |
| 2827 | smp_read_barrier_depends(); |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2828 | if (cic->key == cfqd) |
Jens Axboe | 62c1fe9 | 2008-12-15 21:19:25 +0100 | [diff] [blame] | 2829 | __cfq_exit_single_io_context(cfqd, cic); |
| 2830 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2831 | spin_unlock_irqrestore(q->queue_lock, flags); |
Al Viro | 12a0573 | 2006-03-18 13:38:01 -0500 | [diff] [blame] | 2832 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2833 | } |
| 2834 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 2835 | /* |
| 2836 | * The process that ioc belongs to has exited, we need to clean up |
| 2837 | * and put the internal structures we have that belongs to that process. |
| 2838 | */ |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 2839 | static void cfq_exit_io_context(struct io_context *ioc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2840 | { |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2841 | call_for_each_cic(ioc, cfq_exit_single_io_context); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2842 | } |
| 2843 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2844 | static struct cfq_io_context * |
Al Viro | 8267e26 | 2005-10-21 03:20:53 -0400 | [diff] [blame] | 2845 | cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2846 | { |
Jens Axboe | b5deef9 | 2006-07-19 23:39:40 +0200 | [diff] [blame] | 2847 | struct cfq_io_context *cic; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2848 | |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 2849 | cic = kmem_cache_alloc_node(cfq_ioc_pool, gfp_mask | __GFP_ZERO, |
| 2850 | cfqd->queue->node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2851 | if (cic) { |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 2852 | cic->ttime.last_end_request = jiffies; |
Jens Axboe | 553698f | 2006-06-14 19:11:57 +0200 | [diff] [blame] | 2853 | INIT_LIST_HEAD(&cic->queue_list); |
Jens Axboe | ffc4e75 | 2008-02-19 10:02:29 +0100 | [diff] [blame] | 2854 | INIT_HLIST_NODE(&cic->cic_list); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2855 | cic->dtor = cfq_free_io_context; |
| 2856 | cic->exit = cfq_exit_io_context; |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 2857 | elv_ioc_count_inc(cfq_ioc_count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2858 | } |
| 2859 | |
| 2860 | return cic; |
| 2861 | } |
| 2862 | |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 2863 | static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2864 | { |
| 2865 | struct task_struct *tsk = current; |
| 2866 | int ioprio_class; |
| 2867 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2868 | if (!cfq_cfqq_prio_changed(cfqq)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2869 | return; |
| 2870 | |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 2871 | ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2872 | switch (ioprio_class) { |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 2873 | default: |
| 2874 | printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class); |
| 2875 | case IOPRIO_CLASS_NONE: |
| 2876 | /* |
Jens Axboe | 6d63c27 | 2008-05-07 09:51:23 +0200 | [diff] [blame] | 2877 | * no prio set, inherit CPU scheduling settings |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 2878 | */ |
| 2879 | cfqq->ioprio = task_nice_ioprio(tsk); |
Jens Axboe | 6d63c27 | 2008-05-07 09:51:23 +0200 | [diff] [blame] | 2880 | cfqq->ioprio_class = task_nice_ioclass(tsk); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 2881 | break; |
| 2882 | case IOPRIO_CLASS_RT: |
| 2883 | cfqq->ioprio = task_ioprio(ioc); |
| 2884 | cfqq->ioprio_class = IOPRIO_CLASS_RT; |
| 2885 | break; |
| 2886 | case IOPRIO_CLASS_BE: |
| 2887 | cfqq->ioprio = task_ioprio(ioc); |
| 2888 | cfqq->ioprio_class = IOPRIO_CLASS_BE; |
| 2889 | break; |
| 2890 | case IOPRIO_CLASS_IDLE: |
| 2891 | cfqq->ioprio_class = IOPRIO_CLASS_IDLE; |
| 2892 | cfqq->ioprio = 7; |
| 2893 | cfq_clear_cfqq_idle_window(cfqq); |
| 2894 | break; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2895 | } |
| 2896 | |
| 2897 | /* |
| 2898 | * keep track of original prio settings in case we have to temporarily |
| 2899 | * elevate the priority of this queue |
| 2900 | */ |
| 2901 | cfqq->org_ioprio = cfqq->ioprio; |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 2902 | cfq_clear_cfqq_prio_changed(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2903 | } |
| 2904 | |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 2905 | static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2906 | { |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2907 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
Al Viro | 478a82b | 2006-03-18 13:25:24 -0500 | [diff] [blame] | 2908 | struct cfq_queue *cfqq; |
Jens Axboe | c1b707d | 2006-10-30 19:54:23 +0100 | [diff] [blame] | 2909 | unsigned long flags; |
Jens Axboe | 35e6077 | 2006-06-14 09:10:45 +0200 | [diff] [blame] | 2910 | |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2911 | if (unlikely(!cfqd)) |
| 2912 | return; |
| 2913 | |
Jens Axboe | c1b707d | 2006-10-30 19:54:23 +0100 | [diff] [blame] | 2914 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2915 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 2916 | cfqq = cic->cfqq[BLK_RW_ASYNC]; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2917 | if (cfqq) { |
| 2918 | struct cfq_queue *new_cfqq; |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 2919 | new_cfqq = cfq_get_queue(cfqd, BLK_RW_ASYNC, cic->ioc, |
| 2920 | GFP_ATOMIC); |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2921 | if (new_cfqq) { |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 2922 | cic->cfqq[BLK_RW_ASYNC] = new_cfqq; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2923 | cfq_put_queue(cfqq); |
| 2924 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2925 | } |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2926 | |
Jens Axboe | ff6657c | 2009-04-08 10:58:57 +0200 | [diff] [blame] | 2927 | cfqq = cic->cfqq[BLK_RW_SYNC]; |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 2928 | if (cfqq) |
| 2929 | cfq_mark_cfqq_prio_changed(cfqq); |
| 2930 | |
Jens Axboe | c1b707d | 2006-10-30 19:54:23 +0100 | [diff] [blame] | 2931 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2932 | } |
| 2933 | |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 2934 | static void cfq_ioc_set_ioprio(struct io_context *ioc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2935 | { |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 2936 | call_for_each_cic(ioc, changed_ioprio); |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 2937 | ioc->ioprio_changed = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2938 | } |
| 2939 | |
Jens Axboe | d5036d7 | 2009-06-26 10:44:34 +0200 | [diff] [blame] | 2940 | 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] | 2941 | pid_t pid, bool is_sync) |
Jens Axboe | d5036d7 | 2009-06-26 10:44:34 +0200 | [diff] [blame] | 2942 | { |
| 2943 | RB_CLEAR_NODE(&cfqq->rb_node); |
| 2944 | RB_CLEAR_NODE(&cfqq->p_node); |
| 2945 | INIT_LIST_HEAD(&cfqq->fifo); |
| 2946 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 2947 | cfqq->ref = 0; |
Jens Axboe | d5036d7 | 2009-06-26 10:44:34 +0200 | [diff] [blame] | 2948 | cfqq->cfqd = cfqd; |
| 2949 | |
| 2950 | cfq_mark_cfqq_prio_changed(cfqq); |
| 2951 | |
| 2952 | if (is_sync) { |
| 2953 | if (!cfq_class_idle(cfqq)) |
| 2954 | cfq_mark_cfqq_idle_window(cfqq); |
| 2955 | cfq_mark_cfqq_sync(cfqq); |
| 2956 | } |
| 2957 | cfqq->pid = pid; |
| 2958 | } |
| 2959 | |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 2960 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 2961 | static void changed_cgroup(struct io_context *ioc, struct cfq_io_context *cic) |
| 2962 | { |
| 2963 | struct cfq_queue *sync_cfqq = cic_to_cfqq(cic, 1); |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 2964 | struct cfq_data *cfqd = cic_to_cfqd(cic); |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 2965 | unsigned long flags; |
| 2966 | struct request_queue *q; |
| 2967 | |
| 2968 | if (unlikely(!cfqd)) |
| 2969 | return; |
| 2970 | |
| 2971 | q = cfqd->queue; |
| 2972 | |
| 2973 | spin_lock_irqsave(q->queue_lock, flags); |
| 2974 | |
| 2975 | if (sync_cfqq) { |
| 2976 | /* |
| 2977 | * Drop reference to sync queue. A new sync queue will be |
| 2978 | * assigned in new group upon arrival of a fresh request. |
| 2979 | */ |
| 2980 | cfq_log_cfqq(cfqd, sync_cfqq, "changed cgroup"); |
| 2981 | cic_set_cfqq(cic, NULL, 1); |
| 2982 | cfq_put_queue(sync_cfqq); |
| 2983 | } |
| 2984 | |
| 2985 | spin_unlock_irqrestore(q->queue_lock, flags); |
| 2986 | } |
| 2987 | |
| 2988 | static void cfq_ioc_set_cgroup(struct io_context *ioc) |
| 2989 | { |
| 2990 | call_for_each_cic(ioc, changed_cgroup); |
| 2991 | ioc->cgroup_changed = 0; |
| 2992 | } |
| 2993 | #endif /* CONFIG_CFQ_GROUP_IOSCHED */ |
| 2994 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2995 | static struct cfq_queue * |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 2996 | cfq_find_alloc_queue(struct cfq_data *cfqd, bool is_sync, |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 2997 | struct io_context *ioc, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2998 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2999 | struct cfq_queue *cfqq, *new_cfqq = NULL; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3000 | struct cfq_io_context *cic; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 3001 | struct cfq_group *cfqg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3002 | |
| 3003 | retry: |
Vivek Goyal | 3e59cf9 | 2011-05-19 15:38:21 -0400 | [diff] [blame] | 3004 | cfqg = cfq_get_cfqg(cfqd); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3005 | cic = cfq_cic_lookup(cfqd, ioc); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3006 | /* cic always exists here */ |
| 3007 | cfqq = cic_to_cfqq(cic, is_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3008 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3009 | /* |
| 3010 | * Always try a new alloc if we fell back to the OOM cfqq |
| 3011 | * originally, since it should just be a temporary situation. |
| 3012 | */ |
| 3013 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
| 3014 | cfqq = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3015 | if (new_cfqq) { |
| 3016 | cfqq = new_cfqq; |
| 3017 | new_cfqq = NULL; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3018 | } else if (gfp_mask & __GFP_WAIT) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3019 | spin_unlock_irq(cfqd->queue->queue_lock); |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 3020 | new_cfqq = kmem_cache_alloc_node(cfq_pool, |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3021 | gfp_mask | __GFP_ZERO, |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 3022 | cfqd->queue->node); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3023 | spin_lock_irq(cfqd->queue->queue_lock); |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3024 | if (new_cfqq) |
| 3025 | goto retry; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3026 | } else { |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 3027 | cfqq = kmem_cache_alloc_node(cfq_pool, |
| 3028 | gfp_mask | __GFP_ZERO, |
| 3029 | cfqd->queue->node); |
Kiyoshi Ueda | db3b584 | 2005-06-17 16:15:10 +0200 | [diff] [blame] | 3030 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3032 | if (cfqq) { |
| 3033 | cfq_init_cfqq(cfqd, cfqq, current->pid, is_sync); |
| 3034 | cfq_init_prio_data(cfqq, ioc); |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 3035 | cfq_link_cfqq_cfqg(cfqq, cfqg); |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3036 | cfq_log_cfqq(cfqd, cfqq, "alloced"); |
| 3037 | } else |
| 3038 | cfqq = &cfqd->oom_cfqq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3039 | } |
| 3040 | |
| 3041 | if (new_cfqq) |
| 3042 | kmem_cache_free(cfq_pool, new_cfqq); |
| 3043 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3044 | return cfqq; |
| 3045 | } |
| 3046 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3047 | static struct cfq_queue ** |
| 3048 | cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio) |
| 3049 | { |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3050 | switch (ioprio_class) { |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3051 | case IOPRIO_CLASS_RT: |
| 3052 | return &cfqd->async_cfqq[0][ioprio]; |
| 3053 | case IOPRIO_CLASS_BE: |
| 3054 | return &cfqd->async_cfqq[1][ioprio]; |
| 3055 | case IOPRIO_CLASS_IDLE: |
| 3056 | return &cfqd->async_idle_cfqq; |
| 3057 | default: |
| 3058 | BUG(); |
| 3059 | } |
| 3060 | } |
| 3061 | |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3062 | static struct cfq_queue * |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3063 | cfq_get_queue(struct cfq_data *cfqd, bool is_sync, struct io_context *ioc, |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3064 | gfp_t gfp_mask) |
| 3065 | { |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 3066 | const int ioprio = task_ioprio(ioc); |
| 3067 | const int ioprio_class = task_ioprio_class(ioc); |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3068 | struct cfq_queue **async_cfqq = NULL; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3069 | struct cfq_queue *cfqq = NULL; |
| 3070 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3071 | if (!is_sync) { |
| 3072 | async_cfqq = cfq_async_queue_prio(cfqd, ioprio_class, ioprio); |
| 3073 | cfqq = *async_cfqq; |
| 3074 | } |
| 3075 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 3076 | if (!cfqq) |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 3077 | cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask); |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3078 | |
| 3079 | /* |
| 3080 | * pin the queue now that it's allocated, scheduler exit will prune it |
| 3081 | */ |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3082 | if (!is_sync && !(*async_cfqq)) { |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 3083 | cfqq->ref++; |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3084 | *async_cfqq = cfqq; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3085 | } |
| 3086 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 3087 | cfqq->ref++; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3088 | return cfqq; |
| 3089 | } |
| 3090 | |
Jens Axboe | 498d3aa2 | 2007-04-26 12:54:48 +0200 | [diff] [blame] | 3091 | /* |
| 3092 | * We drop cfq io contexts lazily, so we may find a dead one. |
| 3093 | */ |
OGAWA Hirofumi | dbecf3a | 2006-04-18 09:45:18 +0200 | [diff] [blame] | 3094 | static void |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3095 | cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc, |
| 3096 | struct cfq_io_context *cic) |
OGAWA Hirofumi | dbecf3a | 2006-04-18 09:45:18 +0200 | [diff] [blame] | 3097 | { |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3098 | unsigned long flags; |
| 3099 | |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 3100 | WARN_ON(!list_empty(&cic->queue_list)); |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 3101 | BUG_ON(cic->key != cfqd_dead_key(cfqd)); |
Jens Axboe | 597bc48 | 2007-04-24 21:23:53 +0200 | [diff] [blame] | 3102 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3103 | spin_lock_irqsave(&ioc->lock, flags); |
Jens Axboe | 597bc48 | 2007-04-24 21:23:53 +0200 | [diff] [blame] | 3104 | |
Shaohua Li | 726e99a | 2011-06-27 09:03:48 +0200 | [diff] [blame] | 3105 | BUG_ON(rcu_dereference_check(ioc->ioc_data, |
| 3106 | lockdep_is_held(&ioc->lock)) == cic); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3107 | |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 3108 | radix_tree_delete(&ioc->radix_root, cfqd->cic_index); |
Jens Axboe | ffc4e75 | 2008-02-19 10:02:29 +0100 | [diff] [blame] | 3109 | hlist_del_rcu(&cic->cic_list); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3110 | spin_unlock_irqrestore(&ioc->lock, flags); |
| 3111 | |
| 3112 | cfq_cic_free(cic); |
OGAWA Hirofumi | dbecf3a | 2006-04-18 09:45:18 +0200 | [diff] [blame] | 3113 | } |
| 3114 | |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3115 | static struct cfq_io_context * |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3116 | cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc) |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3117 | { |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3118 | struct cfq_io_context *cic; |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 3119 | unsigned long flags; |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3120 | |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3121 | if (unlikely(!ioc)) |
| 3122 | return NULL; |
| 3123 | |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 3124 | rcu_read_lock(); |
| 3125 | |
Jens Axboe | 597bc48 | 2007-04-24 21:23:53 +0200 | [diff] [blame] | 3126 | /* |
| 3127 | * we maintain a last-hit cache, to avoid browsing over the tree |
| 3128 | */ |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3129 | cic = rcu_dereference(ioc->ioc_data); |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 3130 | if (cic && cic->key == cfqd) { |
| 3131 | rcu_read_unlock(); |
Jens Axboe | 597bc48 | 2007-04-24 21:23:53 +0200 | [diff] [blame] | 3132 | return cic; |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 3133 | } |
Jens Axboe | 597bc48 | 2007-04-24 21:23:53 +0200 | [diff] [blame] | 3134 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3135 | do { |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 3136 | cic = radix_tree_lookup(&ioc->radix_root, cfqd->cic_index); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3137 | rcu_read_unlock(); |
| 3138 | if (!cic) |
| 3139 | break; |
Konstantin Khlebnikov | bca4b91 | 2010-05-20 23:21:34 +0400 | [diff] [blame] | 3140 | if (unlikely(cic->key != cfqd)) { |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3141 | cfq_drop_dead_cic(cfqd, ioc, cic); |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 3142 | rcu_read_lock(); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3143 | continue; |
OGAWA Hirofumi | dbecf3a | 2006-04-18 09:45:18 +0200 | [diff] [blame] | 3144 | } |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3145 | |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 3146 | spin_lock_irqsave(&ioc->lock, flags); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3147 | rcu_assign_pointer(ioc->ioc_data, cic); |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 3148 | spin_unlock_irqrestore(&ioc->lock, flags); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3149 | break; |
| 3150 | } while (1); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3151 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3152 | return cic; |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3153 | } |
| 3154 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3155 | /* |
| 3156 | * Add cic into ioc, using cfqd as the search key. This enables us to lookup |
| 3157 | * the process specific cfq io context when entered from the block layer. |
| 3158 | * Also adds the cic to a per-cfqd list, used when this queue is removed. |
| 3159 | */ |
Jens Axboe | febffd6 | 2008-01-28 13:19:43 +0100 | [diff] [blame] | 3160 | static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc, |
| 3161 | struct cfq_io_context *cic, gfp_t gfp_mask) |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3162 | { |
Jens Axboe | 0261d68 | 2006-10-30 19:07:48 +0100 | [diff] [blame] | 3163 | unsigned long flags; |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3164 | int ret; |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3165 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3166 | ret = radix_tree_preload(gfp_mask); |
| 3167 | if (!ret) { |
| 3168 | cic->ioc = ioc; |
| 3169 | cic->key = cfqd; |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3170 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3171 | spin_lock_irqsave(&ioc->lock, flags); |
| 3172 | ret = radix_tree_insert(&ioc->radix_root, |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 3173 | cfqd->cic_index, cic); |
Jens Axboe | ffc4e75 | 2008-02-19 10:02:29 +0100 | [diff] [blame] | 3174 | if (!ret) |
| 3175 | hlist_add_head_rcu(&cic->cic_list, &ioc->cic_list); |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3176 | spin_unlock_irqrestore(&ioc->lock, flags); |
| 3177 | |
| 3178 | radix_tree_preload_end(); |
| 3179 | |
| 3180 | if (!ret) { |
| 3181 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
| 3182 | list_add(&cic->queue_list, &cfqd->cic_list); |
| 3183 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
OGAWA Hirofumi | dbecf3a | 2006-04-18 09:45:18 +0200 | [diff] [blame] | 3184 | } |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3185 | } |
| 3186 | |
Yasuaki Ishimatsu | 5eb4685 | 2011-12-02 10:07:07 +0100 | [diff] [blame] | 3187 | if (ret && ret != -EEXIST) |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3188 | printk(KERN_ERR "cfq: cic link failed!\n"); |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 3189 | |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3190 | return ret; |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3191 | } |
| 3192 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3193 | /* |
| 3194 | * Setup general io context and cfq io context. There can be several cfq |
| 3195 | * io contexts per general io context, if this process is doing io to more |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3196 | * than one device managed by cfq. |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3197 | */ |
| 3198 | static struct cfq_io_context * |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3199 | cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3200 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3201 | struct io_context *ioc = NULL; |
| 3202 | struct cfq_io_context *cic; |
Yasuaki Ishimatsu | 5eb4685 | 2011-12-02 10:07:07 +0100 | [diff] [blame] | 3203 | int ret; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3204 | |
| 3205 | might_sleep_if(gfp_mask & __GFP_WAIT); |
| 3206 | |
Jens Axboe | b5deef9 | 2006-07-19 23:39:40 +0200 | [diff] [blame] | 3207 | ioc = get_io_context(gfp_mask, cfqd->queue->node); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3208 | if (!ioc) |
| 3209 | return NULL; |
| 3210 | |
Yasuaki Ishimatsu | 5eb4685 | 2011-12-02 10:07:07 +0100 | [diff] [blame] | 3211 | retry: |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3212 | cic = cfq_cic_lookup(cfqd, ioc); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3213 | if (cic) |
| 3214 | goto out; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3215 | |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3216 | cic = cfq_alloc_io_context(cfqd, gfp_mask); |
| 3217 | if (cic == NULL) |
| 3218 | goto err; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3219 | |
Yasuaki Ishimatsu | 5eb4685 | 2011-12-02 10:07:07 +0100 | [diff] [blame] | 3220 | ret = cfq_cic_link(cfqd, ioc, cic, gfp_mask); |
| 3221 | if (ret == -EEXIST) { |
| 3222 | /* someone has linked cic to ioc already */ |
| 3223 | cfq_cic_free(cic); |
| 3224 | goto retry; |
| 3225 | } else if (ret) |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3226 | goto err_free; |
| 3227 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3228 | out: |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 3229 | smp_read_barrier_depends(); |
| 3230 | if (unlikely(ioc->ioprio_changed)) |
| 3231 | cfq_ioc_set_ioprio(ioc); |
| 3232 | |
Vivek Goyal | 24610333 | 2009-12-03 12:59:51 -0500 | [diff] [blame] | 3233 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3234 | if (unlikely(ioc->cgroup_changed)) |
| 3235 | cfq_ioc_set_cgroup(ioc); |
| 3236 | #endif |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3237 | return cic; |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3238 | err_free: |
| 3239 | cfq_cic_free(cic); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3240 | err: |
| 3241 | put_io_context(ioc); |
| 3242 | return NULL; |
| 3243 | } |
| 3244 | |
| 3245 | static void |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3246 | __cfq_update_io_thinktime(struct cfq_ttime *ttime, unsigned long slice_idle) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3247 | { |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3248 | unsigned long elapsed = jiffies - ttime->last_end_request; |
| 3249 | elapsed = min(elapsed, 2UL * slice_idle); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3250 | |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3251 | ttime->ttime_samples = (7*ttime->ttime_samples + 256) / 8; |
| 3252 | ttime->ttime_total = (7*ttime->ttime_total + 256*elapsed) / 8; |
| 3253 | ttime->ttime_mean = (ttime->ttime_total + 128) / ttime->ttime_samples; |
| 3254 | } |
| 3255 | |
| 3256 | static void |
| 3257 | cfq_update_io_thinktime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3258 | struct cfq_io_context *cic) |
| 3259 | { |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3260 | if (cfq_cfqq_sync(cfqq)) { |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3261 | __cfq_update_io_thinktime(&cic->ttime, cfqd->cfq_slice_idle); |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3262 | __cfq_update_io_thinktime(&cfqq->service_tree->ttime, |
| 3263 | cfqd->cfq_slice_idle); |
| 3264 | } |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 3265 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3266 | __cfq_update_io_thinktime(&cfqq->cfqg->ttime, cfqd->cfq_group_idle); |
| 3267 | #endif |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3268 | } |
| 3269 | |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3270 | static void |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 3271 | cfq_update_io_seektime(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3272 | struct request *rq) |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3273 | { |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 3274 | sector_t sdist = 0; |
Corrado Zoccolo | 41647e7 | 2010-02-27 19:45:40 +0100 | [diff] [blame] | 3275 | sector_t n_sec = blk_rq_sectors(rq); |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 3276 | if (cfqq->last_request_pos) { |
| 3277 | if (cfqq->last_request_pos < blk_rq_pos(rq)) |
| 3278 | sdist = blk_rq_pos(rq) - cfqq->last_request_pos; |
| 3279 | else |
| 3280 | sdist = cfqq->last_request_pos - blk_rq_pos(rq); |
| 3281 | } |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3282 | |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 3283 | cfqq->seek_history <<= 1; |
Corrado Zoccolo | 41647e7 | 2010-02-27 19:45:40 +0100 | [diff] [blame] | 3284 | if (blk_queue_nonrot(cfqd->queue)) |
| 3285 | cfqq->seek_history |= (n_sec < CFQQ_SECT_THR_NONROT); |
| 3286 | else |
| 3287 | cfqq->seek_history |= (sdist > CFQQ_SEEK_THR); |
Jens Axboe | 206dc69 | 2006-03-28 13:03:44 +0200 | [diff] [blame] | 3288 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3289 | |
| 3290 | /* |
| 3291 | * Disable idle window if the process thinks too long or seeks so much that |
| 3292 | * it doesn't matter |
| 3293 | */ |
| 3294 | static void |
| 3295 | cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3296 | struct cfq_io_context *cic) |
| 3297 | { |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3298 | int old_idle, enable_idle; |
Jens Axboe | 1be92f2 | 2007-04-19 14:32:26 +0200 | [diff] [blame] | 3299 | |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 3300 | /* |
| 3301 | * Don't idle for async or idle io prio class |
| 3302 | */ |
| 3303 | if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq)) |
Jens Axboe | 1be92f2 | 2007-04-19 14:32:26 +0200 | [diff] [blame] | 3304 | return; |
| 3305 | |
Jens Axboe | c265a7f | 2008-06-26 13:49:33 +0200 | [diff] [blame] | 3306 | enable_idle = old_idle = cfq_cfqq_idle_window(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3307 | |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3308 | if (cfqq->queued[0] + cfqq->queued[1] >= 4) |
| 3309 | cfq_mark_cfqq_deep(cfqq); |
| 3310 | |
Corrado Zoccolo | 749ef9f | 2010-09-20 15:24:50 +0200 | [diff] [blame] | 3311 | if (cfqq->next_rq && (cfqq->next_rq->cmd_flags & REQ_NOIDLE)) |
| 3312 | enable_idle = 0; |
| 3313 | else if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle || |
Corrado Zoccolo | 3dde36d | 2010-02-27 19:45:39 +0100 | [diff] [blame] | 3314 | (!cfq_cfqq_deep(cfqq) && CFQQ_SEEKY(cfqq))) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3315 | enable_idle = 0; |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3316 | else if (sample_valid(cic->ttime.ttime_samples)) { |
| 3317 | if (cic->ttime.ttime_mean > cfqd->cfq_slice_idle) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3318 | enable_idle = 0; |
| 3319 | else |
| 3320 | enable_idle = 1; |
| 3321 | } |
| 3322 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3323 | if (old_idle != enable_idle) { |
| 3324 | cfq_log_cfqq(cfqd, cfqq, "idle=%d", enable_idle); |
| 3325 | if (enable_idle) |
| 3326 | cfq_mark_cfqq_idle_window(cfqq); |
| 3327 | else |
| 3328 | cfq_clear_cfqq_idle_window(cfqq); |
| 3329 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3330 | } |
| 3331 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3332 | /* |
| 3333 | * Check if new_cfqq should preempt the currently active queue. Return 0 for |
| 3334 | * no or if we aren't sure, a 1 will cause a preempt. |
| 3335 | */ |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3336 | static bool |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3337 | cfq_should_preempt(struct cfq_data *cfqd, struct cfq_queue *new_cfqq, |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3338 | struct request *rq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3339 | { |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3340 | struct cfq_queue *cfqq; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3341 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3342 | cfqq = cfqd->active_queue; |
| 3343 | if (!cfqq) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3344 | return false; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3345 | |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3346 | if (cfq_class_idle(new_cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3347 | return false; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3348 | |
| 3349 | if (cfq_class_idle(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3350 | return true; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3351 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3352 | /* |
Divyesh Shah | 875feb6 | 2010-01-06 18:58:20 -0800 | [diff] [blame] | 3353 | * Don't allow a non-RT request to preempt an ongoing RT cfqq timeslice. |
| 3354 | */ |
| 3355 | if (cfq_class_rt(cfqq) && !cfq_class_rt(new_cfqq)) |
| 3356 | return false; |
| 3357 | |
| 3358 | /* |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 3359 | * if the new request is sync, but the currently running queue is |
| 3360 | * not, let the sync request have priority. |
| 3361 | */ |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3362 | if (rq_is_sync(rq) && !cfq_cfqq_sync(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3363 | return true; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3364 | |
Vivek Goyal | 8682e1f | 2009-12-03 12:59:50 -0500 | [diff] [blame] | 3365 | if (new_cfqq->cfqg != cfqq->cfqg) |
| 3366 | return false; |
| 3367 | |
| 3368 | if (cfq_slice_used(cfqq)) |
| 3369 | return true; |
| 3370 | |
| 3371 | /* Allow preemption only if we are idling on sync-noidle tree */ |
| 3372 | if (cfqd->serving_type == SYNC_NOIDLE_WORKLOAD && |
| 3373 | cfqq_type(new_cfqq) == SYNC_NOIDLE_WORKLOAD && |
| 3374 | new_cfqq->service_tree->count == 2 && |
| 3375 | RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 3376 | return true; |
| 3377 | |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 3378 | /* |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 3379 | * So both queues are sync. Let the new request get disk time if |
| 3380 | * it's a metadata request and the current queue is doing regular IO. |
| 3381 | */ |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 3382 | if ((rq->cmd_flags & REQ_PRIO) && !cfqq->prio_pending) |
Jens Axboe | b53d1ed | 2011-08-19 08:34:48 +0200 | [diff] [blame] | 3383 | return true; |
| 3384 | |
| 3385 | /* |
Divyesh Shah | 3a9a3f6 | 2009-01-30 12:46:41 +0100 | [diff] [blame] | 3386 | * Allow an RT request to pre-empt an ongoing non-RT cfqq timeslice. |
| 3387 | */ |
| 3388 | if (cfq_class_rt(new_cfqq) && !cfq_class_rt(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3389 | return true; |
Divyesh Shah | 3a9a3f6 | 2009-01-30 12:46:41 +0100 | [diff] [blame] | 3390 | |
Shaohua Li | d2d59e1 | 2010-11-08 15:01:03 +0100 | [diff] [blame] | 3391 | /* An idle queue should not be idle now for some reason */ |
| 3392 | if (RB_EMPTY_ROOT(&cfqq->sort_list) && !cfq_should_idle(cfqd, cfqq)) |
| 3393 | return true; |
| 3394 | |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3395 | if (!cfqd->active_cic || !cfq_cfqq_wait_request(cfqq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3396 | return false; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3397 | |
| 3398 | /* |
| 3399 | * if this request is as-good as one we would expect from the |
| 3400 | * current cfqq, let it preempt |
| 3401 | */ |
Shaohua Li | e9ce335 | 2010-03-19 08:03:04 +0100 | [diff] [blame] | 3402 | if (cfq_rq_close(cfqd, cfqq, rq)) |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3403 | return true; |
Jens Axboe | 1e3335d | 2007-02-14 19:59:49 +0100 | [diff] [blame] | 3404 | |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3405 | return false; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3406 | } |
| 3407 | |
| 3408 | /* |
| 3409 | * cfqq preempts the active queue. if we allowed preempt with no slice left, |
| 3410 | * let it have half of its nominal slice. |
| 3411 | */ |
| 3412 | static void cfq_preempt_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3413 | { |
Shaohua Li | f8ae6e3 | 2011-01-14 08:41:02 +0100 | [diff] [blame] | 3414 | struct cfq_queue *old_cfqq = cfqd->active_queue; |
| 3415 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3416 | cfq_log_cfqq(cfqd, cfqq, "preempt"); |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3417 | cfq_slice_expired(cfqd, 1); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3418 | |
Jens Axboe | bf57225 | 2006-07-19 20:29:12 +0200 | [diff] [blame] | 3419 | /* |
Shaohua Li | f8ae6e3 | 2011-01-14 08:41:02 +0100 | [diff] [blame] | 3420 | * workload type is changed, don't save slice, otherwise preempt |
| 3421 | * doesn't happen |
| 3422 | */ |
| 3423 | if (cfqq_type(old_cfqq) != cfqq_type(cfqq)) |
| 3424 | cfqq->cfqg->saved_workload_slice = 0; |
| 3425 | |
| 3426 | /* |
Jens Axboe | bf57225 | 2006-07-19 20:29:12 +0200 | [diff] [blame] | 3427 | * Put the new queue at the front of the of the current list, |
| 3428 | * so we know that it will be selected next. |
| 3429 | */ |
| 3430 | BUG_ON(!cfq_cfqq_on_rr(cfqq)); |
Jens Axboe | edd75ff | 2007-04-19 12:03:34 +0200 | [diff] [blame] | 3431 | |
| 3432 | cfq_service_tree_add(cfqd, cfqq, 1); |
Justin TerAvest | eda5e0c | 2011-03-22 21:26:49 +0100 | [diff] [blame] | 3433 | |
Justin TerAvest | 62a37f6 | 2011-03-23 08:25:44 +0100 | [diff] [blame] | 3434 | cfqq->slice_end = 0; |
| 3435 | cfq_mark_cfqq_slice_new(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3436 | } |
| 3437 | |
| 3438 | /* |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3439 | * 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] | 3440 | * something we should do about it |
| 3441 | */ |
| 3442 | static void |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3443 | cfq_rq_enqueued(struct cfq_data *cfqd, struct cfq_queue *cfqq, |
| 3444 | struct request *rq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3445 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3446 | struct cfq_io_context *cic = RQ_CIC(rq); |
Jens Axboe | 12e9fdd | 2006-06-01 10:09:56 +0200 | [diff] [blame] | 3447 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3448 | cfqd->rq_queued++; |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 3449 | if (rq->cmd_flags & REQ_PRIO) |
| 3450 | cfqq->prio_pending++; |
Jens Axboe | 374f84a | 2006-07-23 01:42:19 +0200 | [diff] [blame] | 3451 | |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3452 | cfq_update_io_thinktime(cfqd, cfqq, cic); |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 3453 | cfq_update_io_seektime(cfqd, cfqq, rq); |
Jens Axboe | 9c2c38a | 2005-08-24 14:57:54 +0200 | [diff] [blame] | 3454 | cfq_update_idle_window(cfqd, cfqq, cic); |
| 3455 | |
Jeff Moyer | b2c18e1 | 2009-10-23 17:14:49 -0400 | [diff] [blame] | 3456 | cfqq->last_request_pos = blk_rq_pos(rq) + blk_rq_sectors(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3457 | |
| 3458 | if (cfqq == cfqd->active_queue) { |
| 3459 | /* |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 3460 | * Remember that we saw a request from this process, but |
| 3461 | * don't start queuing just yet. Otherwise we risk seeing lots |
| 3462 | * of tiny requests, because we disrupt the normal plugging |
Jens Axboe | d6ceb25 | 2009-04-14 14:18:16 +0200 | [diff] [blame] | 3463 | * and merging. If the request is already larger than a single |
| 3464 | * page, let it rip immediately. For that case we assume that |
Jens Axboe | 2d87072 | 2009-04-15 12:12:46 +0200 | [diff] [blame] | 3465 | * merging is already done. Ditto for a busy system that |
| 3466 | * has other work pending, don't risk delaying until the |
| 3467 | * idle timer unplug to continue working. |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3468 | */ |
Jens Axboe | d6ceb25 | 2009-04-14 14:18:16 +0200 | [diff] [blame] | 3469 | if (cfq_cfqq_wait_request(cfqq)) { |
Jens Axboe | 2d87072 | 2009-04-15 12:12:46 +0200 | [diff] [blame] | 3470 | if (blk_rq_bytes(rq) > PAGE_CACHE_SIZE || |
| 3471 | cfqd->busy_queues > 1) { |
Divyesh Shah | 812df48 | 2010-04-08 21:15:35 -0700 | [diff] [blame] | 3472 | cfq_del_timer(cfqd, cfqq); |
Gui Jianfeng | 554554f | 2009-12-10 09:38:39 +0100 | [diff] [blame] | 3473 | cfq_clear_cfqq_wait_request(cfqq); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 3474 | __blk_run_queue(cfqd->queue); |
Divyesh Shah | a11cdaa | 2010-04-13 19:59:17 +0200 | [diff] [blame] | 3475 | } else { |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 3476 | cfq_blkiocg_update_idle_time_stats( |
Divyesh Shah | a11cdaa | 2010-04-13 19:59:17 +0200 | [diff] [blame] | 3477 | &cfqq->cfqg->blkg); |
Vivek Goyal | bf791937 | 2009-12-03 12:59:37 -0500 | [diff] [blame] | 3478 | cfq_mark_cfqq_must_dispatch(cfqq); |
Divyesh Shah | a11cdaa | 2010-04-13 19:59:17 +0200 | [diff] [blame] | 3479 | } |
Jens Axboe | d6ceb25 | 2009-04-14 14:18:16 +0200 | [diff] [blame] | 3480 | } |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3481 | } else if (cfq_should_preempt(cfqd, cfqq, rq)) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3482 | /* |
| 3483 | * not the active queue - expire current slice if it is |
| 3484 | * idle and has expired it's mean thinktime or this new queue |
Divyesh Shah | 3a9a3f6 | 2009-01-30 12:46:41 +0100 | [diff] [blame] | 3485 | * has some old slice time left and is of higher priority or |
| 3486 | * this new queue is RT and the current one is BE |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3487 | */ |
| 3488 | cfq_preempt_queue(cfqd, cfqq); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 3489 | __blk_run_queue(cfqd->queue); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3490 | } |
| 3491 | } |
| 3492 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3493 | static void cfq_insert_request(struct request_queue *q, struct request *rq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3494 | { |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3495 | struct cfq_data *cfqd = q->elevator->elevator_data; |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3496 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3497 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3498 | cfq_log_cfqq(cfqd, cfqq, "insert_request"); |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 3499 | cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3500 | |
Jens Axboe | 30996f4 | 2009-10-05 11:03:39 +0200 | [diff] [blame] | 3501 | 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] | 3502 | list_add_tail(&rq->queuelist, &cfqq->fifo); |
Corrado Zoccolo | aa6f6a3 | 2009-10-26 22:44:33 +0100 | [diff] [blame] | 3503 | cfq_add_rq_rb(rq); |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 3504 | cfq_blkiocg_update_io_add_stats(&(RQ_CFQG(rq))->blkg, |
Divyesh Shah | cdc1184 | 2010-04-08 21:15:10 -0700 | [diff] [blame] | 3505 | &cfqd->serving_group->blkg, rq_data_dir(rq), |
| 3506 | rq_is_sync(rq)); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3507 | cfq_rq_enqueued(cfqd, cfqq, rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3508 | } |
| 3509 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3510 | /* |
| 3511 | * Update hw_tag based on peak queue depth over 50 samples under |
| 3512 | * sufficient load. |
| 3513 | */ |
| 3514 | static void cfq_update_hw_tag(struct cfq_data *cfqd) |
| 3515 | { |
Shaohua Li | 1a1238a | 2009-10-27 08:46:23 +0100 | [diff] [blame] | 3516 | struct cfq_queue *cfqq = cfqd->active_queue; |
| 3517 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3518 | if (cfqd->rq_in_driver > cfqd->hw_tag_est_depth) |
| 3519 | cfqd->hw_tag_est_depth = cfqd->rq_in_driver; |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 3520 | |
| 3521 | if (cfqd->hw_tag == 1) |
| 3522 | return; |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3523 | |
| 3524 | if (cfqd->rq_queued <= CFQ_HW_QUEUE_MIN && |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3525 | cfqd->rq_in_driver <= CFQ_HW_QUEUE_MIN) |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3526 | return; |
| 3527 | |
Shaohua Li | 1a1238a | 2009-10-27 08:46:23 +0100 | [diff] [blame] | 3528 | /* |
| 3529 | * If active queue hasn't enough requests and can idle, cfq might not |
| 3530 | * dispatch sufficient requests to hardware. Don't zero hw_tag in this |
| 3531 | * case |
| 3532 | */ |
| 3533 | if (cfqq && cfq_cfqq_idle_window(cfqq) && |
| 3534 | cfqq->dispatched + cfqq->queued[0] + cfqq->queued[1] < |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3535 | CFQ_HW_QUEUE_MIN && cfqd->rq_in_driver < CFQ_HW_QUEUE_MIN) |
Shaohua Li | 1a1238a | 2009-10-27 08:46:23 +0100 | [diff] [blame] | 3536 | return; |
| 3537 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3538 | if (cfqd->hw_tag_samples++ < 50) |
| 3539 | return; |
| 3540 | |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 3541 | if (cfqd->hw_tag_est_depth >= CFQ_HW_QUEUE_MIN) |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3542 | cfqd->hw_tag = 1; |
| 3543 | else |
| 3544 | cfqd->hw_tag = 0; |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3545 | } |
| 3546 | |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3547 | static bool cfq_should_wait_busy(struct cfq_data *cfqd, struct cfq_queue *cfqq) |
| 3548 | { |
| 3549 | struct cfq_io_context *cic = cfqd->active_cic; |
| 3550 | |
Justin TerAvest | 02a8f01 | 2011-02-09 14:20:03 +0100 | [diff] [blame] | 3551 | /* If the queue already has requests, don't wait */ |
| 3552 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
| 3553 | return false; |
| 3554 | |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3555 | /* If there are other queues in the group, don't wait */ |
| 3556 | if (cfqq->cfqg->nr_cfqq > 1) |
| 3557 | return false; |
| 3558 | |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 3559 | /* the only queue in the group, but think time is big */ |
| 3560 | if (cfq_io_thinktime_big(cfqd, &cfqq->cfqg->ttime, true)) |
| 3561 | return false; |
| 3562 | |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3563 | if (cfq_slice_used(cfqq)) |
| 3564 | return true; |
| 3565 | |
| 3566 | /* if slice left is less than think time, wait busy */ |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3567 | if (cic && sample_valid(cic->ttime.ttime_samples) |
| 3568 | && (cfqq->slice_end - jiffies < cic->ttime.ttime_mean)) |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3569 | return true; |
| 3570 | |
| 3571 | /* |
| 3572 | * If think times is less than a jiffy than ttime_mean=0 and above |
| 3573 | * will not be true. It might happen that slice has not expired yet |
| 3574 | * but will expire soon (4-5 ns) during select_queue(). To cover the |
| 3575 | * case where think time is less than a jiffy, mark the queue wait |
| 3576 | * busy if only 1 jiffy is left in the slice. |
| 3577 | */ |
| 3578 | if (cfqq->slice_end - jiffies == 1) |
| 3579 | return true; |
| 3580 | |
| 3581 | return false; |
| 3582 | } |
| 3583 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3584 | static void cfq_completed_request(struct request_queue *q, struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3585 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3586 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3587 | struct cfq_data *cfqd = cfqq->cfqd; |
Jens Axboe | 5380a10 | 2006-07-13 12:37:56 +0200 | [diff] [blame] | 3588 | const int sync = rq_is_sync(rq); |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3589 | unsigned long now; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3590 | |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 3591 | now = jiffies; |
Christoph Hellwig | 33659eb | 2010-08-07 18:17:56 +0200 | [diff] [blame] | 3592 | cfq_log_cfqq(cfqd, cfqq, "complete rqnoidle %d", |
| 3593 | !!(rq->cmd_flags & REQ_NOIDLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3594 | |
Aaron Carroll | 45333d5 | 2008-08-26 15:52:36 +0200 | [diff] [blame] | 3595 | cfq_update_hw_tag(cfqd); |
| 3596 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3597 | WARN_ON(!cfqd->rq_in_driver); |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3598 | WARN_ON(!cfqq->dispatched); |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3599 | cfqd->rq_in_driver--; |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3600 | cfqq->dispatched--; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 3601 | (RQ_CFQG(rq))->dispatched--; |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 3602 | cfq_blkiocg_update_completion_stats(&cfqq->cfqg->blkg, |
| 3603 | rq_start_time_ns(rq), rq_io_start_time_ns(rq), |
| 3604 | rq_data_dir(rq), rq_is_sync(rq)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3605 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3606 | cfqd->rq_in_flight[cfq_cfqq_sync(cfqq)]--; |
Jens Axboe | 3ed9a29 | 2007-04-23 08:33:33 +0200 | [diff] [blame] | 3607 | |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 3608 | if (sync) { |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3609 | struct cfq_rb_root *service_tree; |
| 3610 | |
Shaohua Li | 383cd72 | 2011-07-12 14:24:35 +0200 | [diff] [blame] | 3611 | RQ_CIC(rq)->ttime.last_end_request = now; |
Shaohua Li | f5f2b6c | 2011-07-12 14:24:55 +0200 | [diff] [blame] | 3612 | |
| 3613 | if (cfq_cfqq_on_rr(cfqq)) |
| 3614 | service_tree = cfqq->service_tree; |
| 3615 | else |
| 3616 | service_tree = service_tree_for(cfqq->cfqg, |
| 3617 | cfqq_prio(cfqq), cfqq_type(cfqq)); |
| 3618 | service_tree->ttime.last_end_request = now; |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 3619 | if (!time_after(rq->start_time + cfqd->cfq_fifo_expire[1], now)) |
| 3620 | cfqd->last_delayed_sync = now; |
Vivek Goyal | 365722b | 2009-10-03 15:21:27 +0200 | [diff] [blame] | 3621 | } |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3622 | |
Shaohua Li | 7700fc4 | 2011-07-12 14:24:56 +0200 | [diff] [blame] | 3623 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3624 | cfqq->cfqg->ttime.last_end_request = now; |
| 3625 | #endif |
| 3626 | |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3627 | /* |
| 3628 | * If this is the active queue, check if it needs to be expired, |
| 3629 | * or if we want to idle in case it has no pending requests. |
| 3630 | */ |
| 3631 | if (cfqd->active_queue == cfqq) { |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 3632 | const bool cfqq_empty = RB_EMPTY_ROOT(&cfqq->sort_list); |
| 3633 | |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 3634 | if (cfq_cfqq_slice_new(cfqq)) { |
| 3635 | cfq_set_prio_slice(cfqd, cfqq); |
| 3636 | cfq_clear_cfqq_slice_new(cfqq); |
| 3637 | } |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3638 | |
| 3639 | /* |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3640 | * Should we wait for next request to come in before we expire |
| 3641 | * the queue. |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3642 | */ |
Vivek Goyal | 7667aa0 | 2009-12-08 17:52:58 -0500 | [diff] [blame] | 3643 | if (cfq_should_wait_busy(cfqd, cfqq)) { |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 3644 | unsigned long extend_sl = cfqd->cfq_slice_idle; |
| 3645 | if (!cfqd->cfq_slice_idle) |
| 3646 | extend_sl = cfqd->cfq_group_idle; |
| 3647 | cfqq->slice_end = jiffies + extend_sl; |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3648 | cfq_mark_cfqq_wait_busy(cfqq); |
Divyesh Shah | b1ffe73 | 2010-03-25 15:45:03 +0100 | [diff] [blame] | 3649 | cfq_log_cfqq(cfqd, cfqq, "will busy wait"); |
Vivek Goyal | f75edf2 | 2009-12-03 12:59:53 -0500 | [diff] [blame] | 3650 | } |
| 3651 | |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 3652 | /* |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3653 | * Idling is not enabled on: |
| 3654 | * - expired queues |
| 3655 | * - idle-priority queues |
| 3656 | * - async queues |
| 3657 | * - queues with still some requests queued |
| 3658 | * - when there is a close cooperator |
Jens Axboe | a36e71f | 2009-04-15 12:15:11 +0200 | [diff] [blame] | 3659 | */ |
Jens Axboe | 0871714 | 2008-01-28 11:38:15 +0100 | [diff] [blame] | 3660 | if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq)) |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3661 | cfq_slice_expired(cfqd, 1); |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3662 | else if (sync && cfqq_empty && |
| 3663 | !cfq_close_cooperator(cfqd, cfqq)) { |
Corrado Zoccolo | 749ef9f | 2010-09-20 15:24:50 +0200 | [diff] [blame] | 3664 | cfq_arm_slice_timer(cfqd); |
Corrado Zoccolo | 8e55063 | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3665 | } |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3666 | } |
Jens Axboe | 6d048f5 | 2007-04-25 12:44:27 +0200 | [diff] [blame] | 3667 | |
Corrado Zoccolo | 53c583d | 2010-02-28 19:45:05 +0100 | [diff] [blame] | 3668 | if (!cfqd->rq_in_driver) |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3669 | cfq_schedule_dispatch(cfqd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3670 | } |
| 3671 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3672 | static inline int __cfq_may_queue(struct cfq_queue *cfqq) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3673 | { |
Jens Axboe | 1b379d8 | 2009-08-11 08:26:11 +0200 | [diff] [blame] | 3674 | if (cfq_cfqq_wait_request(cfqq) && !cfq_cfqq_must_alloc_slice(cfqq)) { |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3675 | cfq_mark_cfqq_must_alloc_slice(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3676 | return ELV_MQUEUE_MUST; |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3677 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3678 | |
| 3679 | return ELV_MQUEUE_MAY; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3680 | } |
| 3681 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3682 | static int cfq_may_queue(struct request_queue *q, int rw) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3683 | { |
| 3684 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 3685 | struct task_struct *tsk = current; |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3686 | struct cfq_io_context *cic; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3687 | struct cfq_queue *cfqq; |
| 3688 | |
| 3689 | /* |
| 3690 | * don't force setup of a queue from here, as a call to may_queue |
| 3691 | * does not necessarily imply that a request actually will be queued. |
| 3692 | * so just lookup a possibly existing queue, or return 'may queue' |
| 3693 | * if that fails |
| 3694 | */ |
Jens Axboe | 4ac845a | 2008-01-24 08:44:49 +0100 | [diff] [blame] | 3695 | cic = cfq_cic_lookup(cfqd, tsk->io_context); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3696 | if (!cic) |
| 3697 | return ELV_MQUEUE_MAY; |
| 3698 | |
Jens Axboe | b0b78f8 | 2009-04-08 10:56:08 +0200 | [diff] [blame] | 3699 | cfqq = cic_to_cfqq(cic, rw_is_sync(rw)); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3700 | if (cfqq) { |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 3701 | cfq_init_prio_data(cfqq, cic->ioc); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3702 | |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3703 | return __cfq_may_queue(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3704 | } |
| 3705 | |
| 3706 | return ELV_MQUEUE_MAY; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3707 | } |
| 3708 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3709 | /* |
| 3710 | * queue lock held here |
| 3711 | */ |
Jens Axboe | bb37b94 | 2006-12-01 10:42:33 +0100 | [diff] [blame] | 3712 | static void cfq_put_request(struct request *rq) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3713 | { |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3714 | struct cfq_queue *cfqq = RQ_CFQQ(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3715 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3716 | if (cfqq) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3717 | const int rw = rq_data_dir(rq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3718 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3719 | BUG_ON(!cfqq->allocated[rw]); |
| 3720 | cfqq->allocated[rw]--; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3721 | |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3722 | put_io_context(RQ_CIC(rq)->ioc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3723 | |
Mike Snitzer | c186794 | 2011-02-11 11:08:00 +0100 | [diff] [blame] | 3724 | rq->elevator_private[0] = NULL; |
| 3725 | rq->elevator_private[1] = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3726 | |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 3727 | /* Put down rq reference on cfqg */ |
| 3728 | cfq_put_cfqg(RQ_CFQG(rq)); |
Mike Snitzer | c186794 | 2011-02-11 11:08:00 +0100 | [diff] [blame] | 3729 | rq->elevator_private[2] = NULL; |
Vivek Goyal | 7f1dc8a | 2010-04-21 17:44:16 +0200 | [diff] [blame] | 3730 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3731 | cfq_put_queue(cfqq); |
| 3732 | } |
| 3733 | } |
| 3734 | |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3735 | static struct cfq_queue * |
| 3736 | cfq_merge_cfqqs(struct cfq_data *cfqd, struct cfq_io_context *cic, |
| 3737 | struct cfq_queue *cfqq) |
| 3738 | { |
| 3739 | cfq_log_cfqq(cfqd, cfqq, "merging with queue %p", cfqq->new_cfqq); |
| 3740 | cic_set_cfqq(cic, cfqq->new_cfqq, 1); |
Jeff Moyer | b3b6d04 | 2009-10-23 17:14:51 -0400 | [diff] [blame] | 3741 | cfq_mark_cfqq_coop(cfqq->new_cfqq); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3742 | cfq_put_queue(cfqq); |
| 3743 | return cic_to_cfqq(cic, 1); |
| 3744 | } |
| 3745 | |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3746 | /* |
| 3747 | * Returns NULL if a new cfqq should be allocated, or the old cfqq if this |
| 3748 | * was the last process referring to said cfqq. |
| 3749 | */ |
| 3750 | static struct cfq_queue * |
| 3751 | split_cfqq(struct cfq_io_context *cic, struct cfq_queue *cfqq) |
| 3752 | { |
| 3753 | if (cfqq_process_refs(cfqq) == 1) { |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3754 | cfqq->pid = current->pid; |
| 3755 | cfq_clear_cfqq_coop(cfqq); |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 3756 | cfq_clear_cfqq_split_coop(cfqq); |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3757 | return cfqq; |
| 3758 | } |
| 3759 | |
| 3760 | cic_set_cfqq(cic, NULL, 1); |
Shaohua Li | d02a2c0 | 2010-05-25 10:16:53 +0200 | [diff] [blame] | 3761 | |
| 3762 | cfq_put_cooperator(cfqq); |
| 3763 | |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3764 | cfq_put_queue(cfqq); |
| 3765 | return NULL; |
| 3766 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3767 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3768 | * Allocate cfq data structures associated with this request. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3769 | */ |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3770 | static int |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3771 | cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3772 | { |
| 3773 | struct cfq_data *cfqd = q->elevator->elevator_data; |
| 3774 | struct cfq_io_context *cic; |
| 3775 | const int rw = rq_data_dir(rq); |
Jens Axboe | a6151c3 | 2009-10-07 20:02:57 +0200 | [diff] [blame] | 3776 | const bool is_sync = rq_is_sync(rq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3777 | struct cfq_queue *cfqq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3778 | unsigned long flags; |
| 3779 | |
| 3780 | might_sleep_if(gfp_mask & __GFP_WAIT); |
| 3781 | |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3782 | cic = cfq_get_io_context(cfqd, gfp_mask); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3783 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3784 | spin_lock_irqsave(q->queue_lock, flags); |
| 3785 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3786 | if (!cic) |
| 3787 | goto queue_fail; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3788 | |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3789 | new_queue: |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3790 | cfqq = cic_to_cfqq(cic, is_sync); |
Vivek Goyal | 32f2e80 | 2009-07-09 22:13:16 +0200 | [diff] [blame] | 3791 | if (!cfqq || cfqq == &cfqd->oom_cfqq) { |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 3792 | cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3793 | cic_set_cfqq(cic, cfqq, is_sync); |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3794 | } else { |
| 3795 | /* |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3796 | * If the queue was seeky for too long, break it apart. |
| 3797 | */ |
Shaohua Li | ae54abe | 2010-02-05 13:11:45 +0100 | [diff] [blame] | 3798 | if (cfq_cfqq_coop(cfqq) && cfq_cfqq_split_coop(cfqq)) { |
Jeff Moyer | e6c5bc7 | 2009-10-23 17:14:52 -0400 | [diff] [blame] | 3799 | cfq_log_cfqq(cfqd, cfqq, "breaking apart cfqq"); |
| 3800 | cfqq = split_cfqq(cic, cfqq); |
| 3801 | if (!cfqq) |
| 3802 | goto new_queue; |
| 3803 | } |
| 3804 | |
| 3805 | /* |
Jeff Moyer | df5fe3e | 2009-10-23 17:14:50 -0400 | [diff] [blame] | 3806 | * Check to see if this queue is scheduled to merge with |
| 3807 | * another, closely cooperating queue. The merging of |
| 3808 | * queues happens here as it must be done in process context. |
| 3809 | * The reference on new_cfqq was taken in merge_cfqqs. |
| 3810 | */ |
| 3811 | if (cfqq->new_cfqq) |
| 3812 | cfqq = cfq_merge_cfqqs(cfqd, cic, cfqq); |
Vasily Tarasov | 91fac31 | 2007-04-25 12:29:51 +0200 | [diff] [blame] | 3813 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3814 | |
| 3815 | cfqq->allocated[rw]++; |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3816 | |
Jens Axboe | 6fae9c2 | 2011-03-01 15:04:39 -0500 | [diff] [blame] | 3817 | cfqq->ref++; |
Mike Snitzer | c186794 | 2011-02-11 11:08:00 +0100 | [diff] [blame] | 3818 | rq->elevator_private[0] = cic; |
| 3819 | rq->elevator_private[1] = cfqq; |
| 3820 | rq->elevator_private[2] = cfq_ref_get_cfqg(cfqq->cfqg); |
Jens Axboe | 93803e0 | 2011-03-07 08:59:06 +0100 | [diff] [blame] | 3821 | spin_unlock_irqrestore(q->queue_lock, flags); |
Jens Axboe | 5e70537 | 2006-07-13 12:39:25 +0200 | [diff] [blame] | 3822 | return 0; |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3823 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3824 | queue_fail: |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3825 | cfq_schedule_dispatch(cfqd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3826 | spin_unlock_irqrestore(q->queue_lock, flags); |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3827 | cfq_log(cfqd, "set_request fail"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3828 | return 1; |
| 3829 | } |
| 3830 | |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3831 | static void cfq_kick_queue(struct work_struct *work) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3832 | { |
David Howells | 65f27f3 | 2006-11-22 14:55:48 +0000 | [diff] [blame] | 3833 | struct cfq_data *cfqd = |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3834 | container_of(work, struct cfq_data, unplug_work); |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3835 | struct request_queue *q = cfqd->queue; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3836 | |
Jens Axboe | 40bb54d | 2009-04-15 12:11:10 +0200 | [diff] [blame] | 3837 | spin_lock_irq(q->queue_lock); |
Christoph Hellwig | 24ecfbe | 2011-04-18 11:41:33 +0200 | [diff] [blame] | 3838 | __blk_run_queue(cfqd->queue); |
Jens Axboe | 40bb54d | 2009-04-15 12:11:10 +0200 | [diff] [blame] | 3839 | spin_unlock_irq(q->queue_lock); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3840 | } |
| 3841 | |
| 3842 | /* |
| 3843 | * Timer running if the active_queue is currently idling inside its time slice |
| 3844 | */ |
| 3845 | static void cfq_idle_slice_timer(unsigned long data) |
| 3846 | { |
| 3847 | struct cfq_data *cfqd = (struct cfq_data *) data; |
| 3848 | struct cfq_queue *cfqq; |
| 3849 | unsigned long flags; |
Jens Axboe | 3c6bd2f | 2007-01-19 12:06:33 +1100 | [diff] [blame] | 3850 | int timed_out = 1; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3851 | |
Jens Axboe | 7b67913 | 2008-05-30 12:23:07 +0200 | [diff] [blame] | 3852 | cfq_log(cfqd, "idle timer fired"); |
| 3853 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3854 | spin_lock_irqsave(cfqd->queue->queue_lock, flags); |
| 3855 | |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 3856 | cfqq = cfqd->active_queue; |
| 3857 | if (cfqq) { |
Jens Axboe | 3c6bd2f | 2007-01-19 12:06:33 +1100 | [diff] [blame] | 3858 | timed_out = 0; |
| 3859 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3860 | /* |
Jens Axboe | b029195 | 2009-04-07 11:38:31 +0200 | [diff] [blame] | 3861 | * We saw a request before the queue expired, let it through |
| 3862 | */ |
| 3863 | if (cfq_cfqq_must_dispatch(cfqq)) |
| 3864 | goto out_kick; |
| 3865 | |
| 3866 | /* |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3867 | * expired |
| 3868 | */ |
Jens Axboe | 44f7c16 | 2007-01-19 11:51:58 +1100 | [diff] [blame] | 3869 | if (cfq_slice_used(cfqq)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3870 | goto expire; |
| 3871 | |
| 3872 | /* |
| 3873 | * only expire and reinvoke request handler, if there are |
| 3874 | * other queues with pending requests |
| 3875 | */ |
Jens Axboe | caaa5f9 | 2006-06-16 11:23:00 +0200 | [diff] [blame] | 3876 | if (!cfqd->busy_queues) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3877 | goto out_cont; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3878 | |
| 3879 | /* |
| 3880 | * not expired and it has a request pending, let it dispatch |
| 3881 | */ |
Jens Axboe | 75e5098 | 2009-04-07 08:56:14 +0200 | [diff] [blame] | 3882 | if (!RB_EMPTY_ROOT(&cfqq->sort_list)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3883 | goto out_kick; |
Corrado Zoccolo | 76280af | 2009-11-26 10:02:58 +0100 | [diff] [blame] | 3884 | |
| 3885 | /* |
| 3886 | * Queue depth flag is reset only when the idle didn't succeed |
| 3887 | */ |
| 3888 | cfq_clear_cfqq_deep(cfqq); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3889 | } |
| 3890 | expire: |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3891 | cfq_slice_expired(cfqd, timed_out); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3892 | out_kick: |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3893 | cfq_schedule_dispatch(cfqd); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3894 | out_cont: |
| 3895 | spin_unlock_irqrestore(cfqd->queue->queue_lock, flags); |
| 3896 | } |
| 3897 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3898 | static void cfq_shutdown_timer_wq(struct cfq_data *cfqd) |
| 3899 | { |
| 3900 | del_timer_sync(&cfqd->idle_slice_timer); |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 3901 | cancel_work_sync(&cfqd->unplug_work); |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3902 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3903 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3904 | static void cfq_put_async_queues(struct cfq_data *cfqd) |
| 3905 | { |
| 3906 | int i; |
| 3907 | |
| 3908 | for (i = 0; i < IOPRIO_BE_NR; i++) { |
| 3909 | if (cfqd->async_cfqq[0][i]) |
| 3910 | cfq_put_queue(cfqd->async_cfqq[0][i]); |
| 3911 | if (cfqd->async_cfqq[1][i]) |
| 3912 | cfq_put_queue(cfqd->async_cfqq[1][i]); |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3913 | } |
Oleg Nesterov | 2389d1e | 2007-11-05 08:58:05 +0100 | [diff] [blame] | 3914 | |
| 3915 | if (cfqd->async_idle_cfqq) |
| 3916 | cfq_put_queue(cfqd->async_idle_cfqq); |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3917 | } |
| 3918 | |
Jens Axboe | b374d18 | 2008-10-31 10:05:07 +0100 | [diff] [blame] | 3919 | static void cfq_exit_queue(struct elevator_queue *e) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3920 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3921 | struct cfq_data *cfqd = e->elevator_data; |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3922 | struct request_queue *q = cfqd->queue; |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 3923 | bool wait = false; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 3924 | |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 3925 | cfq_shutdown_timer_wq(cfqd); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3926 | |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 3927 | spin_lock_irq(q->queue_lock); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3928 | |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 3929 | if (cfqd->active_queue) |
Vivek Goyal | e5ff082 | 2010-04-26 19:25:11 +0200 | [diff] [blame] | 3930 | __cfq_slice_expired(cfqd, cfqd->active_queue, 0); |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3931 | |
| 3932 | while (!list_empty(&cfqd->cic_list)) { |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 3933 | struct cfq_io_context *cic = list_entry(cfqd->cic_list.next, |
| 3934 | struct cfq_io_context, |
| 3935 | queue_list); |
Jens Axboe | 89850f7 | 2006-07-22 16:48:31 +0200 | [diff] [blame] | 3936 | |
| 3937 | __cfq_exit_single_io_context(cfqd, cic); |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 3938 | } |
Jens Axboe | e2d74ac | 2006-03-28 08:59:01 +0200 | [diff] [blame] | 3939 | |
Vasily Tarasov | c2dea2d | 2007-07-20 10:06:38 +0200 | [diff] [blame] | 3940 | cfq_put_async_queues(cfqd); |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 3941 | cfq_release_cfq_groups(cfqd); |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 3942 | |
| 3943 | /* |
| 3944 | * If there are groups which we could not unlink from blkcg list, |
| 3945 | * wait for a rcu period for them to be freed. |
| 3946 | */ |
| 3947 | if (cfqd->nr_blkcg_linked_grps) |
| 3948 | wait = true; |
Jens Axboe | 15c31be | 2007-07-10 13:43:25 +0200 | [diff] [blame] | 3949 | |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 3950 | spin_unlock_irq(q->queue_lock); |
Al Viro | a90d742 | 2006-03-18 12:05:37 -0500 | [diff] [blame] | 3951 | |
| 3952 | cfq_shutdown_timer_wq(cfqd); |
| 3953 | |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 3954 | spin_lock(&cic_index_lock); |
| 3955 | ida_remove(&cic_index_ida, cfqd->cic_index); |
| 3956 | spin_unlock(&cic_index_lock); |
| 3957 | |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 3958 | /* |
| 3959 | * Wait for cfqg->blkg->key accessors to exit their grace periods. |
| 3960 | * Do this wait only if there are other unlinked groups out |
| 3961 | * there. This can happen if cgroup deletion path claimed the |
| 3962 | * responsibility of cleaning up a group before queue cleanup code |
| 3963 | * get to the group. |
| 3964 | * |
| 3965 | * Do not call synchronize_rcu() unconditionally as there are drivers |
| 3966 | * which create/delete request queue hundreds of times during scan/boot |
| 3967 | * and synchronize_rcu() can take significant time and slow down boot. |
| 3968 | */ |
| 3969 | if (wait) |
| 3970 | synchronize_rcu(); |
Vivek Goyal | 2abae55 | 2011-05-23 10:02:19 +0200 | [diff] [blame] | 3971 | |
| 3972 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 3973 | /* Free up per cpu stats for root group */ |
| 3974 | free_percpu(cfqd->root_group.blkg.stats_cpu); |
| 3975 | #endif |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 3976 | kfree(cfqd); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3977 | } |
| 3978 | |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 3979 | static int cfq_alloc_cic_index(void) |
| 3980 | { |
| 3981 | int index, error; |
| 3982 | |
| 3983 | do { |
| 3984 | if (!ida_pre_get(&cic_index_ida, GFP_KERNEL)) |
| 3985 | return -ENOMEM; |
| 3986 | |
| 3987 | spin_lock(&cic_index_lock); |
| 3988 | error = ida_get_new(&cic_index_ida, &index); |
| 3989 | spin_unlock(&cic_index_lock); |
| 3990 | if (error && error != -EAGAIN) |
| 3991 | return error; |
| 3992 | } while (error); |
| 3993 | |
| 3994 | return index; |
| 3995 | } |
| 3996 | |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 3997 | static void *cfq_init_queue(struct request_queue *q) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3998 | { |
| 3999 | struct cfq_data *cfqd; |
Corrado Zoccolo | 718eee0 | 2009-10-26 22:45:29 +0100 | [diff] [blame] | 4000 | int i, j; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 4001 | struct cfq_group *cfqg; |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 4002 | struct cfq_rb_root *st; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4003 | |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 4004 | i = cfq_alloc_cic_index(); |
| 4005 | if (i < 0) |
| 4006 | return NULL; |
| 4007 | |
Christoph Lameter | 94f6030 | 2007-07-17 04:03:29 -0700 | [diff] [blame] | 4008 | cfqd = kmalloc_node(sizeof(*cfqd), GFP_KERNEL | __GFP_ZERO, q->node); |
Namhyung Kim | 1547010 | 2011-05-24 10:23:22 +0200 | [diff] [blame] | 4009 | if (!cfqd) { |
| 4010 | spin_lock(&cic_index_lock); |
| 4011 | ida_remove(&cic_index_ida, i); |
| 4012 | spin_unlock(&cic_index_lock); |
Jens Axboe | bc1c116 | 2006-06-08 08:49:06 +0200 | [diff] [blame] | 4013 | return NULL; |
Namhyung Kim | 1547010 | 2011-05-24 10:23:22 +0200 | [diff] [blame] | 4014 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4015 | |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 4016 | /* |
| 4017 | * Don't need take queue_lock in the routine, since we are |
| 4018 | * initializing the ioscheduler, and nobody is using cfqd |
| 4019 | */ |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 4020 | cfqd->cic_index = i; |
| 4021 | |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 4022 | /* Init root service tree */ |
| 4023 | cfqd->grp_service_tree = CFQ_RB_ROOT; |
| 4024 | |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 4025 | /* Init root group */ |
| 4026 | cfqg = &cfqd->root_group; |
Vivek Goyal | 615f025 | 2009-12-03 12:59:39 -0500 | [diff] [blame] | 4027 | for_each_cfqg_st(cfqg, i, j, st) |
| 4028 | *st = CFQ_RB_ROOT; |
Vivek Goyal | 1fa8f6d | 2009-12-03 12:59:41 -0500 | [diff] [blame] | 4029 | RB_CLEAR_NODE(&cfqg->rb_node); |
Jens Axboe | 26a2ac0 | 2009-04-23 12:13:27 +0200 | [diff] [blame] | 4030 | |
Vivek Goyal | 25bc6b0 | 2009-12-03 12:59:43 -0500 | [diff] [blame] | 4031 | /* Give preference to root group over other groups */ |
| 4032 | cfqg->weight = 2*BLKIO_WEIGHT_DEFAULT; |
| 4033 | |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 4034 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 4035 | /* |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 4036 | * Set root group reference to 2. One reference will be dropped when |
| 4037 | * all groups on cfqd->cfqg_list are being deleted during queue exit. |
| 4038 | * Other reference will remain there as we don't want to delete this |
| 4039 | * group as it is statically allocated and gets destroyed when |
| 4040 | * throtl_data goes away. |
Vivek Goyal | b1c3576 | 2009-12-03 12:59:47 -0500 | [diff] [blame] | 4041 | */ |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 4042 | cfqg->ref = 2; |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 4043 | |
| 4044 | if (blkio_alloc_blkg_stats(&cfqg->blkg)) { |
| 4045 | kfree(cfqg); |
majianpeng | 2984ff3 | 2011-11-30 15:47:48 +0100 | [diff] [blame] | 4046 | |
| 4047 | spin_lock(&cic_index_lock); |
| 4048 | ida_remove(&cic_index_ida, cfqd->cic_index); |
| 4049 | spin_unlock(&cic_index_lock); |
| 4050 | |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 4051 | kfree(cfqd); |
| 4052 | return NULL; |
| 4053 | } |
| 4054 | |
Vivek Goyal | dcf097b | 2010-04-22 11:54:52 -0400 | [diff] [blame] | 4055 | rcu_read_lock(); |
Vivek Goyal | 5624a4e | 2011-05-19 15:38:28 -0400 | [diff] [blame] | 4056 | |
Vivek Goyal | e98ef89 | 2010-06-18 10:39:47 -0400 | [diff] [blame] | 4057 | cfq_blkiocg_add_blkio_group(&blkio_root_cgroup, &cfqg->blkg, |
| 4058 | (void *)cfqd, 0); |
Vivek Goyal | dcf097b | 2010-04-22 11:54:52 -0400 | [diff] [blame] | 4059 | rcu_read_unlock(); |
Vivek Goyal | 56edf7d | 2011-05-19 15:38:22 -0400 | [diff] [blame] | 4060 | cfqd->nr_blkcg_linked_grps++; |
| 4061 | |
| 4062 | /* Add group on cfqd->cfqg_list */ |
| 4063 | hlist_add_head(&cfqg->cfqd_node, &cfqd->cfqg_list); |
Vivek Goyal | 25fb516 | 2009-12-03 12:59:46 -0500 | [diff] [blame] | 4064 | #endif |
Jens Axboe | 26a2ac0 | 2009-04-23 12:13:27 +0200 | [diff] [blame] | 4065 | /* |
| 4066 | * Not strictly needed (since RB_ROOT just clears the node and we |
| 4067 | * zeroed cfqd on alloc), but better be safe in case someone decides |
| 4068 | * to add magic to the rb code |
| 4069 | */ |
| 4070 | for (i = 0; i < CFQ_PRIO_LISTS; i++) |
| 4071 | cfqd->prio_trees[i] = RB_ROOT; |
| 4072 | |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 4073 | /* |
| 4074 | * Our fallback cfqq if cfq_find_alloc_queue() runs into OOM issues. |
| 4075 | * Grab a permanent reference to it, so that the normal code flow |
| 4076 | * will not attempt to free it. |
| 4077 | */ |
| 4078 | cfq_init_cfqq(cfqd, &cfqd->oom_cfqq, 1, 0); |
Shaohua Li | 30d7b94 | 2011-01-07 08:46:59 +0100 | [diff] [blame] | 4079 | cfqd->oom_cfqq.ref++; |
Vivek Goyal | cdb16e8 | 2009-12-03 12:59:38 -0500 | [diff] [blame] | 4080 | cfq_link_cfqq_cfqg(&cfqd->oom_cfqq, &cfqd->root_group); |
Jens Axboe | 6118b70 | 2009-06-30 09:34:12 +0200 | [diff] [blame] | 4081 | |
Al Viro | d9ff418 | 2006-03-18 13:51:22 -0500 | [diff] [blame] | 4082 | INIT_LIST_HEAD(&cfqd->cic_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4083 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4084 | cfqd->queue = q; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4085 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4086 | init_timer(&cfqd->idle_slice_timer); |
| 4087 | cfqd->idle_slice_timer.function = cfq_idle_slice_timer; |
| 4088 | cfqd->idle_slice_timer.data = (unsigned long) cfqd; |
| 4089 | |
Jens Axboe | 23e018a | 2009-10-05 08:52:35 +0200 | [diff] [blame] | 4090 | INIT_WORK(&cfqd->unplug_work, cfq_kick_queue); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4091 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4092 | cfqd->cfq_quantum = cfq_quantum; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4093 | cfqd->cfq_fifo_expire[0] = cfq_fifo_expire[0]; |
| 4094 | cfqd->cfq_fifo_expire[1] = cfq_fifo_expire[1]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4095 | cfqd->cfq_back_max = cfq_back_max; |
| 4096 | cfqd->cfq_back_penalty = cfq_back_penalty; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4097 | cfqd->cfq_slice[0] = cfq_slice_async; |
| 4098 | cfqd->cfq_slice[1] = cfq_slice_sync; |
| 4099 | cfqd->cfq_slice_async_rq = cfq_slice_async_rq; |
| 4100 | cfqd->cfq_slice_idle = cfq_slice_idle; |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4101 | cfqd->cfq_group_idle = cfq_group_idle; |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 4102 | cfqd->cfq_latency = 1; |
Corrado Zoccolo | e459dd0 | 2009-11-26 10:02:57 +0100 | [diff] [blame] | 4103 | cfqd->hw_tag = -1; |
Corrado Zoccolo | edc7113 | 2009-12-09 20:56:04 +0100 | [diff] [blame] | 4104 | /* |
| 4105 | * we optimistically start assuming sync ops weren't delayed in last |
| 4106 | * second, in order to have larger depth for async operations. |
| 4107 | */ |
Corrado Zoccolo | 573412b | 2009-12-06 11:48:52 +0100 | [diff] [blame] | 4108 | cfqd->last_delayed_sync = jiffies - HZ; |
Jens Axboe | bc1c116 | 2006-06-08 08:49:06 +0200 | [diff] [blame] | 4109 | return cfqd; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4110 | } |
| 4111 | |
| 4112 | static void cfq_slab_kill(void) |
| 4113 | { |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 4114 | /* |
| 4115 | * Caller already ensured that pending RCU callbacks are completed, |
| 4116 | * so we should have no busy allocations at this point. |
| 4117 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4118 | if (cfq_pool) |
| 4119 | kmem_cache_destroy(cfq_pool); |
| 4120 | if (cfq_ioc_pool) |
| 4121 | kmem_cache_destroy(cfq_ioc_pool); |
| 4122 | } |
| 4123 | |
| 4124 | static int __init cfq_slab_setup(void) |
| 4125 | { |
Christoph Lameter | 0a31bd5 | 2007-05-06 14:49:57 -0700 | [diff] [blame] | 4126 | cfq_pool = KMEM_CACHE(cfq_queue, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4127 | if (!cfq_pool) |
| 4128 | goto fail; |
| 4129 | |
Fabio Checconi | 34e6bbf | 2008-04-02 14:31:02 +0200 | [diff] [blame] | 4130 | cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4131 | if (!cfq_ioc_pool) |
| 4132 | goto fail; |
| 4133 | |
| 4134 | return 0; |
| 4135 | fail: |
| 4136 | cfq_slab_kill(); |
| 4137 | return -ENOMEM; |
| 4138 | } |
| 4139 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4140 | /* |
| 4141 | * sysfs parts below --> |
| 4142 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4143 | static ssize_t |
| 4144 | cfq_var_show(unsigned int var, char *page) |
| 4145 | { |
| 4146 | return sprintf(page, "%d\n", var); |
| 4147 | } |
| 4148 | |
| 4149 | static ssize_t |
| 4150 | cfq_var_store(unsigned int *var, const char *page, size_t count) |
| 4151 | { |
| 4152 | char *p = (char *) page; |
| 4153 | |
| 4154 | *var = simple_strtoul(p, &p, 10); |
| 4155 | return count; |
| 4156 | } |
| 4157 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4158 | #define SHOW_FUNCTION(__FUNC, __VAR, __CONV) \ |
Jens Axboe | b374d18 | 2008-10-31 10:05:07 +0100 | [diff] [blame] | 4159 | static ssize_t __FUNC(struct elevator_queue *e, char *page) \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4160 | { \ |
Al Viro | 3d1ab40 | 2006-03-18 18:35:43 -0500 | [diff] [blame] | 4161 | struct cfq_data *cfqd = e->elevator_data; \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4162 | unsigned int __data = __VAR; \ |
| 4163 | if (__CONV) \ |
| 4164 | __data = jiffies_to_msecs(__data); \ |
| 4165 | return cfq_var_show(__data, (page)); \ |
| 4166 | } |
| 4167 | SHOW_FUNCTION(cfq_quantum_show, cfqd->cfq_quantum, 0); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4168 | SHOW_FUNCTION(cfq_fifo_expire_sync_show, cfqd->cfq_fifo_expire[1], 1); |
| 4169 | 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] | 4170 | SHOW_FUNCTION(cfq_back_seek_max_show, cfqd->cfq_back_max, 0); |
| 4171 | SHOW_FUNCTION(cfq_back_seek_penalty_show, cfqd->cfq_back_penalty, 0); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4172 | SHOW_FUNCTION(cfq_slice_idle_show, cfqd->cfq_slice_idle, 1); |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4173 | SHOW_FUNCTION(cfq_group_idle_show, cfqd->cfq_group_idle, 1); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4174 | SHOW_FUNCTION(cfq_slice_sync_show, cfqd->cfq_slice[1], 1); |
| 4175 | SHOW_FUNCTION(cfq_slice_async_show, cfqd->cfq_slice[0], 1); |
| 4176 | 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] | 4177 | SHOW_FUNCTION(cfq_low_latency_show, cfqd->cfq_latency, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4178 | #undef SHOW_FUNCTION |
| 4179 | |
| 4180 | #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV) \ |
Jens Axboe | b374d18 | 2008-10-31 10:05:07 +0100 | [diff] [blame] | 4181 | 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] | 4182 | { \ |
Al Viro | 3d1ab40 | 2006-03-18 18:35:43 -0500 | [diff] [blame] | 4183 | struct cfq_data *cfqd = e->elevator_data; \ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4184 | unsigned int __data; \ |
| 4185 | int ret = cfq_var_store(&__data, (page), count); \ |
| 4186 | if (__data < (MIN)) \ |
| 4187 | __data = (MIN); \ |
| 4188 | else if (__data > (MAX)) \ |
| 4189 | __data = (MAX); \ |
| 4190 | if (__CONV) \ |
| 4191 | *(__PTR) = msecs_to_jiffies(__data); \ |
| 4192 | else \ |
| 4193 | *(__PTR) = __data; \ |
| 4194 | return ret; \ |
| 4195 | } |
| 4196 | STORE_FUNCTION(cfq_quantum_store, &cfqd->cfq_quantum, 1, UINT_MAX, 0); |
Jens Axboe | fe094d9 | 2008-01-31 13:08:54 +0100 | [diff] [blame] | 4197 | STORE_FUNCTION(cfq_fifo_expire_sync_store, &cfqd->cfq_fifo_expire[1], 1, |
| 4198 | UINT_MAX, 1); |
| 4199 | STORE_FUNCTION(cfq_fifo_expire_async_store, &cfqd->cfq_fifo_expire[0], 1, |
| 4200 | UINT_MAX, 1); |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4201 | 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] | 4202 | STORE_FUNCTION(cfq_back_seek_penalty_store, &cfqd->cfq_back_penalty, 1, |
| 4203 | UINT_MAX, 0); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4204 | 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] | 4205 | 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] | 4206 | STORE_FUNCTION(cfq_slice_sync_store, &cfqd->cfq_slice[1], 1, UINT_MAX, 1); |
| 4207 | 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] | 4208 | STORE_FUNCTION(cfq_slice_async_rq_store, &cfqd->cfq_slice_async_rq, 1, |
| 4209 | UINT_MAX, 0); |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 4210 | STORE_FUNCTION(cfq_low_latency_store, &cfqd->cfq_latency, 0, 1, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4211 | #undef STORE_FUNCTION |
| 4212 | |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4213 | #define CFQ_ATTR(name) \ |
| 4214 | __ATTR(name, S_IRUGO|S_IWUSR, cfq_##name##_show, cfq_##name##_store) |
Jens Axboe | 3b18152 | 2005-06-27 10:56:24 +0200 | [diff] [blame] | 4215 | |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4216 | static struct elv_fs_entry cfq_attrs[] = { |
| 4217 | CFQ_ATTR(quantum), |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4218 | CFQ_ATTR(fifo_expire_sync), |
| 4219 | CFQ_ATTR(fifo_expire_async), |
| 4220 | CFQ_ATTR(back_seek_max), |
| 4221 | CFQ_ATTR(back_seek_penalty), |
| 4222 | CFQ_ATTR(slice_sync), |
| 4223 | CFQ_ATTR(slice_async), |
| 4224 | CFQ_ATTR(slice_async_rq), |
| 4225 | CFQ_ATTR(slice_idle), |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4226 | CFQ_ATTR(group_idle), |
Jens Axboe | 963b72f | 2009-10-03 19:42:18 +0200 | [diff] [blame] | 4227 | CFQ_ATTR(low_latency), |
Al Viro | e572ec7 | 2006-03-18 22:27:18 -0500 | [diff] [blame] | 4228 | __ATTR_NULL |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4229 | }; |
| 4230 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4231 | static struct elevator_type iosched_cfq = { |
| 4232 | .ops = { |
| 4233 | .elevator_merge_fn = cfq_merge, |
| 4234 | .elevator_merged_fn = cfq_merged_request, |
| 4235 | .elevator_merge_req_fn = cfq_merged_requests, |
Jens Axboe | da77526 | 2006-12-20 11:04:12 +0100 | [diff] [blame] | 4236 | .elevator_allow_merge_fn = cfq_allow_merge, |
Divyesh Shah | 812d402 | 2010-04-08 21:14:23 -0700 | [diff] [blame] | 4237 | .elevator_bio_merged_fn = cfq_bio_merged, |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 4238 | .elevator_dispatch_fn = cfq_dispatch_requests, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4239 | .elevator_add_req_fn = cfq_insert_request, |
Jens Axboe | b4878f2 | 2005-10-20 16:42:29 +0200 | [diff] [blame] | 4240 | .elevator_activate_req_fn = cfq_activate_request, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4241 | .elevator_deactivate_req_fn = cfq_deactivate_request, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4242 | .elevator_completed_req_fn = cfq_completed_request, |
Jens Axboe | 21183b0 | 2006-07-13 12:33:14 +0200 | [diff] [blame] | 4243 | .elevator_former_req_fn = elv_rb_former_request, |
| 4244 | .elevator_latter_req_fn = elv_rb_latter_request, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4245 | .elevator_set_req_fn = cfq_set_request, |
| 4246 | .elevator_put_req_fn = cfq_put_request, |
| 4247 | .elevator_may_queue_fn = cfq_may_queue, |
| 4248 | .elevator_init_fn = cfq_init_queue, |
| 4249 | .elevator_exit_fn = cfq_exit_queue, |
Jens Axboe | fc46379 | 2006-08-29 09:05:44 +0200 | [diff] [blame] | 4250 | .trim = cfq_free_io_context, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4251 | }, |
Al Viro | 3d1ab40 | 2006-03-18 18:35:43 -0500 | [diff] [blame] | 4252 | .elevator_attrs = cfq_attrs, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4253 | .elevator_name = "cfq", |
| 4254 | .elevator_owner = THIS_MODULE, |
| 4255 | }; |
| 4256 | |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4257 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4258 | static struct blkio_policy_type blkio_policy_cfq = { |
| 4259 | .ops = { |
| 4260 | .blkio_unlink_group_fn = cfq_unlink_blkio_group, |
| 4261 | .blkio_update_group_weight_fn = cfq_update_blkio_group_weight, |
| 4262 | }, |
Vivek Goyal | 062a644 | 2010-09-15 17:06:33 -0400 | [diff] [blame] | 4263 | .plid = BLKIO_POLICY_PROP, |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4264 | }; |
| 4265 | #else |
| 4266 | static struct blkio_policy_type blkio_policy_cfq; |
| 4267 | #endif |
| 4268 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4269 | static int __init cfq_init(void) |
| 4270 | { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 4271 | /* |
| 4272 | * could be 0 on HZ < 1000 setups |
| 4273 | */ |
| 4274 | if (!cfq_slice_async) |
| 4275 | cfq_slice_async = 1; |
| 4276 | if (!cfq_slice_idle) |
| 4277 | cfq_slice_idle = 1; |
| 4278 | |
Vivek Goyal | 80bdf0c | 2010-08-23 12:24:26 +0200 | [diff] [blame] | 4279 | #ifdef CONFIG_CFQ_GROUP_IOSCHED |
| 4280 | if (!cfq_group_idle) |
| 4281 | cfq_group_idle = 1; |
| 4282 | #else |
| 4283 | cfq_group_idle = 0; |
| 4284 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4285 | if (cfq_slab_setup()) |
| 4286 | return -ENOMEM; |
| 4287 | |
Adrian Bunk | 2fdd82b | 2007-12-12 18:51:56 +0100 | [diff] [blame] | 4288 | elv_register(&iosched_cfq); |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4289 | blkio_policy_register(&blkio_policy_cfq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4290 | |
Adrian Bunk | 2fdd82b | 2007-12-12 18:51:56 +0100 | [diff] [blame] | 4291 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4292 | } |
| 4293 | |
| 4294 | static void __exit cfq_exit(void) |
| 4295 | { |
Peter Zijlstra | 6e9a473 | 2006-09-30 23:28:10 -0700 | [diff] [blame] | 4296 | DECLARE_COMPLETION_ONSTACK(all_gone); |
Vivek Goyal | 3e25206 | 2009-12-04 10:36:42 -0500 | [diff] [blame] | 4297 | blkio_policy_unregister(&blkio_policy_cfq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4298 | elv_unregister(&iosched_cfq); |
Al Viro | 334e94d | 2006-03-18 15:05:53 -0500 | [diff] [blame] | 4299 | ioc_gone = &all_gone; |
OGAWA Hirofumi | fba8227 | 2006-04-18 09:44:06 +0200 | [diff] [blame] | 4300 | /* ioc_gone's update must be visible before reading ioc_count */ |
| 4301 | smp_wmb(); |
Jens Axboe | d6de8be | 2008-05-28 14:46:59 +0200 | [diff] [blame] | 4302 | |
| 4303 | /* |
| 4304 | * this also protects us from entering cfq_slab_kill() with |
| 4305 | * pending RCU callbacks |
| 4306 | */ |
Tejun Heo | 245b2e7 | 2009-06-24 15:13:48 +0900 | [diff] [blame] | 4307 | if (elv_ioc_count_read(cfq_ioc_count)) |
Jens Axboe | 9a11b4e | 2008-05-29 09:32:08 +0200 | [diff] [blame] | 4308 | wait_for_completion(&all_gone); |
Konstantin Khlebnikov | 80b15c7 | 2010-05-20 23:21:41 +0400 | [diff] [blame] | 4309 | ida_destroy(&cic_index_ida); |
Christoph Hellwig | 83521d3 | 2005-10-30 15:01:39 -0800 | [diff] [blame] | 4310 | cfq_slab_kill(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4311 | } |
| 4312 | |
| 4313 | module_init(cfq_init); |
| 4314 | module_exit(cfq_exit); |
| 4315 | |
| 4316 | MODULE_AUTHOR("Jens Axboe"); |
| 4317 | MODULE_LICENSE("GPL"); |
| 4318 | MODULE_DESCRIPTION("Completely Fair Queueing IO scheduler"); |