Commit 88ee5ef157202624de2b43b3512fdcb54fda1ab5
Committed by
Jens Axboe
1 parent
ef9be1d336
[BLOCK] ll_rw_blk: fastpath get_request()
Originally from: Nick Piggin <nickpiggin@yahoo.com.au> Move current_io_context out of the get_request fastpth. Also try to streamline a few other things in this area. Signed-off-by: Jens Axboe <axboe@suse.de>
Showing 1 changed file with 37 additions and 33 deletions Inline Diff
block/ll_rw_blk.c
1 | /* | 1 | /* |
2 | * Copyright (C) 1991, 1992 Linus Torvalds | 2 | * Copyright (C) 1991, 1992 Linus Torvalds |
3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics | 3 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics |
4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | 4 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE |
5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> | 5 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> |
6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000 | 6 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> - July2000 |
7 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 | 7 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
8 | */ | 8 | */ |
9 | 9 | ||
10 | /* | 10 | /* |
11 | * This handles all read/write requests to block devices | 11 | * This handles all read/write requests to block devices |
12 | */ | 12 | */ |
13 | #include <linux/config.h> | 13 | #include <linux/config.h> |
14 | #include <linux/kernel.h> | 14 | #include <linux/kernel.h> |
15 | #include <linux/module.h> | 15 | #include <linux/module.h> |
16 | #include <linux/backing-dev.h> | 16 | #include <linux/backing-dev.h> |
17 | #include <linux/bio.h> | 17 | #include <linux/bio.h> |
18 | #include <linux/blkdev.h> | 18 | #include <linux/blkdev.h> |
19 | #include <linux/highmem.h> | 19 | #include <linux/highmem.h> |
20 | #include <linux/mm.h> | 20 | #include <linux/mm.h> |
21 | #include <linux/kernel_stat.h> | 21 | #include <linux/kernel_stat.h> |
22 | #include <linux/string.h> | 22 | #include <linux/string.h> |
23 | #include <linux/init.h> | 23 | #include <linux/init.h> |
24 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ | 24 | #include <linux/bootmem.h> /* for max_pfn/max_low_pfn */ |
25 | #include <linux/completion.h> | 25 | #include <linux/completion.h> |
26 | #include <linux/slab.h> | 26 | #include <linux/slab.h> |
27 | #include <linux/swap.h> | 27 | #include <linux/swap.h> |
28 | #include <linux/writeback.h> | 28 | #include <linux/writeback.h> |
29 | #include <linux/blkdev.h> | 29 | #include <linux/blkdev.h> |
30 | 30 | ||
31 | /* | 31 | /* |
32 | * for max sense size | 32 | * for max sense size |
33 | */ | 33 | */ |
34 | #include <scsi/scsi_cmnd.h> | 34 | #include <scsi/scsi_cmnd.h> |
35 | 35 | ||
36 | static void blk_unplug_work(void *data); | 36 | static void blk_unplug_work(void *data); |
37 | static void blk_unplug_timeout(unsigned long data); | 37 | static void blk_unplug_timeout(unsigned long data); |
38 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io); | 38 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io); |
39 | 39 | ||
40 | /* | 40 | /* |
41 | * For the allocated request tables | 41 | * For the allocated request tables |
42 | */ | 42 | */ |
43 | static kmem_cache_t *request_cachep; | 43 | static kmem_cache_t *request_cachep; |
44 | 44 | ||
45 | /* | 45 | /* |
46 | * For queue allocation | 46 | * For queue allocation |
47 | */ | 47 | */ |
48 | static kmem_cache_t *requestq_cachep; | 48 | static kmem_cache_t *requestq_cachep; |
49 | 49 | ||
50 | /* | 50 | /* |
51 | * For io context allocations | 51 | * For io context allocations |
52 | */ | 52 | */ |
53 | static kmem_cache_t *iocontext_cachep; | 53 | static kmem_cache_t *iocontext_cachep; |
54 | 54 | ||
55 | static wait_queue_head_t congestion_wqh[2] = { | 55 | static wait_queue_head_t congestion_wqh[2] = { |
56 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | 56 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), |
57 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | 57 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) |
58 | }; | 58 | }; |
59 | 59 | ||
60 | /* | 60 | /* |
61 | * Controlling structure to kblockd | 61 | * Controlling structure to kblockd |
62 | */ | 62 | */ |
63 | static struct workqueue_struct *kblockd_workqueue; | 63 | static struct workqueue_struct *kblockd_workqueue; |
64 | 64 | ||
65 | unsigned long blk_max_low_pfn, blk_max_pfn; | 65 | unsigned long blk_max_low_pfn, blk_max_pfn; |
66 | 66 | ||
67 | EXPORT_SYMBOL(blk_max_low_pfn); | 67 | EXPORT_SYMBOL(blk_max_low_pfn); |
68 | EXPORT_SYMBOL(blk_max_pfn); | 68 | EXPORT_SYMBOL(blk_max_pfn); |
69 | 69 | ||
70 | /* Amount of time in which a process may batch requests */ | 70 | /* Amount of time in which a process may batch requests */ |
71 | #define BLK_BATCH_TIME (HZ/50UL) | 71 | #define BLK_BATCH_TIME (HZ/50UL) |
72 | 72 | ||
73 | /* Number of requests a "batching" process may submit */ | 73 | /* Number of requests a "batching" process may submit */ |
74 | #define BLK_BATCH_REQ 32 | 74 | #define BLK_BATCH_REQ 32 |
75 | 75 | ||
76 | /* | 76 | /* |
77 | * Return the threshold (number of used requests) at which the queue is | 77 | * Return the threshold (number of used requests) at which the queue is |
78 | * considered to be congested. It include a little hysteresis to keep the | 78 | * considered to be congested. It include a little hysteresis to keep the |
79 | * context switch rate down. | 79 | * context switch rate down. |
80 | */ | 80 | */ |
81 | static inline int queue_congestion_on_threshold(struct request_queue *q) | 81 | static inline int queue_congestion_on_threshold(struct request_queue *q) |
82 | { | 82 | { |
83 | return q->nr_congestion_on; | 83 | return q->nr_congestion_on; |
84 | } | 84 | } |
85 | 85 | ||
86 | /* | 86 | /* |
87 | * The threshold at which a queue is considered to be uncongested | 87 | * The threshold at which a queue is considered to be uncongested |
88 | */ | 88 | */ |
89 | static inline int queue_congestion_off_threshold(struct request_queue *q) | 89 | static inline int queue_congestion_off_threshold(struct request_queue *q) |
90 | { | 90 | { |
91 | return q->nr_congestion_off; | 91 | return q->nr_congestion_off; |
92 | } | 92 | } |
93 | 93 | ||
94 | static void blk_queue_congestion_threshold(struct request_queue *q) | 94 | static void blk_queue_congestion_threshold(struct request_queue *q) |
95 | { | 95 | { |
96 | int nr; | 96 | int nr; |
97 | 97 | ||
98 | nr = q->nr_requests - (q->nr_requests / 8) + 1; | 98 | nr = q->nr_requests - (q->nr_requests / 8) + 1; |
99 | if (nr > q->nr_requests) | 99 | if (nr > q->nr_requests) |
100 | nr = q->nr_requests; | 100 | nr = q->nr_requests; |
101 | q->nr_congestion_on = nr; | 101 | q->nr_congestion_on = nr; |
102 | 102 | ||
103 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; | 103 | nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; |
104 | if (nr < 1) | 104 | if (nr < 1) |
105 | nr = 1; | 105 | nr = 1; |
106 | q->nr_congestion_off = nr; | 106 | q->nr_congestion_off = nr; |
107 | } | 107 | } |
108 | 108 | ||
109 | /* | 109 | /* |
110 | * A queue has just exitted congestion. Note this in the global counter of | 110 | * A queue has just exitted congestion. Note this in the global counter of |
111 | * congested queues, and wake up anyone who was waiting for requests to be | 111 | * congested queues, and wake up anyone who was waiting for requests to be |
112 | * put back. | 112 | * put back. |
113 | */ | 113 | */ |
114 | static void clear_queue_congested(request_queue_t *q, int rw) | 114 | static void clear_queue_congested(request_queue_t *q, int rw) |
115 | { | 115 | { |
116 | enum bdi_state bit; | 116 | enum bdi_state bit; |
117 | wait_queue_head_t *wqh = &congestion_wqh[rw]; | 117 | wait_queue_head_t *wqh = &congestion_wqh[rw]; |
118 | 118 | ||
119 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; | 119 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; |
120 | clear_bit(bit, &q->backing_dev_info.state); | 120 | clear_bit(bit, &q->backing_dev_info.state); |
121 | smp_mb__after_clear_bit(); | 121 | smp_mb__after_clear_bit(); |
122 | if (waitqueue_active(wqh)) | 122 | if (waitqueue_active(wqh)) |
123 | wake_up(wqh); | 123 | wake_up(wqh); |
124 | } | 124 | } |
125 | 125 | ||
126 | /* | 126 | /* |
127 | * A queue has just entered congestion. Flag that in the queue's VM-visible | 127 | * A queue has just entered congestion. Flag that in the queue's VM-visible |
128 | * state flags and increment the global gounter of congested queues. | 128 | * state flags and increment the global gounter of congested queues. |
129 | */ | 129 | */ |
130 | static void set_queue_congested(request_queue_t *q, int rw) | 130 | static void set_queue_congested(request_queue_t *q, int rw) |
131 | { | 131 | { |
132 | enum bdi_state bit; | 132 | enum bdi_state bit; |
133 | 133 | ||
134 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; | 134 | bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; |
135 | set_bit(bit, &q->backing_dev_info.state); | 135 | set_bit(bit, &q->backing_dev_info.state); |
136 | } | 136 | } |
137 | 137 | ||
138 | /** | 138 | /** |
139 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info | 139 | * blk_get_backing_dev_info - get the address of a queue's backing_dev_info |
140 | * @bdev: device | 140 | * @bdev: device |
141 | * | 141 | * |
142 | * Locates the passed device's request queue and returns the address of its | 142 | * Locates the passed device's request queue and returns the address of its |
143 | * backing_dev_info | 143 | * backing_dev_info |
144 | * | 144 | * |
145 | * Will return NULL if the request queue cannot be located. | 145 | * Will return NULL if the request queue cannot be located. |
146 | */ | 146 | */ |
147 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) | 147 | struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) |
148 | { | 148 | { |
149 | struct backing_dev_info *ret = NULL; | 149 | struct backing_dev_info *ret = NULL; |
150 | request_queue_t *q = bdev_get_queue(bdev); | 150 | request_queue_t *q = bdev_get_queue(bdev); |
151 | 151 | ||
152 | if (q) | 152 | if (q) |
153 | ret = &q->backing_dev_info; | 153 | ret = &q->backing_dev_info; |
154 | return ret; | 154 | return ret; |
155 | } | 155 | } |
156 | 156 | ||
157 | EXPORT_SYMBOL(blk_get_backing_dev_info); | 157 | EXPORT_SYMBOL(blk_get_backing_dev_info); |
158 | 158 | ||
159 | void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data) | 159 | void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data) |
160 | { | 160 | { |
161 | q->activity_fn = fn; | 161 | q->activity_fn = fn; |
162 | q->activity_data = data; | 162 | q->activity_data = data; |
163 | } | 163 | } |
164 | 164 | ||
165 | EXPORT_SYMBOL(blk_queue_activity_fn); | 165 | EXPORT_SYMBOL(blk_queue_activity_fn); |
166 | 166 | ||
167 | /** | 167 | /** |
168 | * blk_queue_prep_rq - set a prepare_request function for queue | 168 | * blk_queue_prep_rq - set a prepare_request function for queue |
169 | * @q: queue | 169 | * @q: queue |
170 | * @pfn: prepare_request function | 170 | * @pfn: prepare_request function |
171 | * | 171 | * |
172 | * It's possible for a queue to register a prepare_request callback which | 172 | * It's possible for a queue to register a prepare_request callback which |
173 | * is invoked before the request is handed to the request_fn. The goal of | 173 | * is invoked before the request is handed to the request_fn. The goal of |
174 | * the function is to prepare a request for I/O, it can be used to build a | 174 | * the function is to prepare a request for I/O, it can be used to build a |
175 | * cdb from the request data for instance. | 175 | * cdb from the request data for instance. |
176 | * | 176 | * |
177 | */ | 177 | */ |
178 | void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn) | 178 | void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn) |
179 | { | 179 | { |
180 | q->prep_rq_fn = pfn; | 180 | q->prep_rq_fn = pfn; |
181 | } | 181 | } |
182 | 182 | ||
183 | EXPORT_SYMBOL(blk_queue_prep_rq); | 183 | EXPORT_SYMBOL(blk_queue_prep_rq); |
184 | 184 | ||
185 | /** | 185 | /** |
186 | * blk_queue_merge_bvec - set a merge_bvec function for queue | 186 | * blk_queue_merge_bvec - set a merge_bvec function for queue |
187 | * @q: queue | 187 | * @q: queue |
188 | * @mbfn: merge_bvec_fn | 188 | * @mbfn: merge_bvec_fn |
189 | * | 189 | * |
190 | * Usually queues have static limitations on the max sectors or segments that | 190 | * Usually queues have static limitations on the max sectors or segments that |
191 | * we can put in a request. Stacking drivers may have some settings that | 191 | * we can put in a request. Stacking drivers may have some settings that |
192 | * are dynamic, and thus we have to query the queue whether it is ok to | 192 | * are dynamic, and thus we have to query the queue whether it is ok to |
193 | * add a new bio_vec to a bio at a given offset or not. If the block device | 193 | * add a new bio_vec to a bio at a given offset or not. If the block device |
194 | * has such limitations, it needs to register a merge_bvec_fn to control | 194 | * has such limitations, it needs to register a merge_bvec_fn to control |
195 | * the size of bio's sent to it. Note that a block device *must* allow a | 195 | * the size of bio's sent to it. Note that a block device *must* allow a |
196 | * single page to be added to an empty bio. The block device driver may want | 196 | * single page to be added to an empty bio. The block device driver may want |
197 | * to use the bio_split() function to deal with these bio's. By default | 197 | * to use the bio_split() function to deal with these bio's. By default |
198 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are | 198 | * no merge_bvec_fn is defined for a queue, and only the fixed limits are |
199 | * honored. | 199 | * honored. |
200 | */ | 200 | */ |
201 | void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn) | 201 | void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn) |
202 | { | 202 | { |
203 | q->merge_bvec_fn = mbfn; | 203 | q->merge_bvec_fn = mbfn; |
204 | } | 204 | } |
205 | 205 | ||
206 | EXPORT_SYMBOL(blk_queue_merge_bvec); | 206 | EXPORT_SYMBOL(blk_queue_merge_bvec); |
207 | 207 | ||
208 | /** | 208 | /** |
209 | * blk_queue_make_request - define an alternate make_request function for a device | 209 | * blk_queue_make_request - define an alternate make_request function for a device |
210 | * @q: the request queue for the device to be affected | 210 | * @q: the request queue for the device to be affected |
211 | * @mfn: the alternate make_request function | 211 | * @mfn: the alternate make_request function |
212 | * | 212 | * |
213 | * Description: | 213 | * Description: |
214 | * The normal way for &struct bios to be passed to a device | 214 | * The normal way for &struct bios to be passed to a device |
215 | * driver is for them to be collected into requests on a request | 215 | * driver is for them to be collected into requests on a request |
216 | * queue, and then to allow the device driver to select requests | 216 | * queue, and then to allow the device driver to select requests |
217 | * off that queue when it is ready. This works well for many block | 217 | * off that queue when it is ready. This works well for many block |
218 | * devices. However some block devices (typically virtual devices | 218 | * devices. However some block devices (typically virtual devices |
219 | * such as md or lvm) do not benefit from the processing on the | 219 | * such as md or lvm) do not benefit from the processing on the |
220 | * request queue, and are served best by having the requests passed | 220 | * request queue, and are served best by having the requests passed |
221 | * directly to them. This can be achieved by providing a function | 221 | * directly to them. This can be achieved by providing a function |
222 | * to blk_queue_make_request(). | 222 | * to blk_queue_make_request(). |
223 | * | 223 | * |
224 | * Caveat: | 224 | * Caveat: |
225 | * The driver that does this *must* be able to deal appropriately | 225 | * The driver that does this *must* be able to deal appropriately |
226 | * with buffers in "highmemory". This can be accomplished by either calling | 226 | * with buffers in "highmemory". This can be accomplished by either calling |
227 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling | 227 | * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling |
228 | * blk_queue_bounce() to create a buffer in normal memory. | 228 | * blk_queue_bounce() to create a buffer in normal memory. |
229 | **/ | 229 | **/ |
230 | void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn) | 230 | void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn) |
231 | { | 231 | { |
232 | /* | 232 | /* |
233 | * set defaults | 233 | * set defaults |
234 | */ | 234 | */ |
235 | q->nr_requests = BLKDEV_MAX_RQ; | 235 | q->nr_requests = BLKDEV_MAX_RQ; |
236 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | 236 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
237 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | 237 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); |
238 | q->make_request_fn = mfn; | 238 | q->make_request_fn = mfn; |
239 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; | 239 | q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; |
240 | q->backing_dev_info.state = 0; | 240 | q->backing_dev_info.state = 0; |
241 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; | 241 | q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; |
242 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); | 242 | blk_queue_max_sectors(q, SAFE_MAX_SECTORS); |
243 | blk_queue_hardsect_size(q, 512); | 243 | blk_queue_hardsect_size(q, 512); |
244 | blk_queue_dma_alignment(q, 511); | 244 | blk_queue_dma_alignment(q, 511); |
245 | blk_queue_congestion_threshold(q); | 245 | blk_queue_congestion_threshold(q); |
246 | q->nr_batching = BLK_BATCH_REQ; | 246 | q->nr_batching = BLK_BATCH_REQ; |
247 | 247 | ||
248 | q->unplug_thresh = 4; /* hmm */ | 248 | q->unplug_thresh = 4; /* hmm */ |
249 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ | 249 | q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ |
250 | if (q->unplug_delay == 0) | 250 | if (q->unplug_delay == 0) |
251 | q->unplug_delay = 1; | 251 | q->unplug_delay = 1; |
252 | 252 | ||
253 | INIT_WORK(&q->unplug_work, blk_unplug_work, q); | 253 | INIT_WORK(&q->unplug_work, blk_unplug_work, q); |
254 | 254 | ||
255 | q->unplug_timer.function = blk_unplug_timeout; | 255 | q->unplug_timer.function = blk_unplug_timeout; |
256 | q->unplug_timer.data = (unsigned long)q; | 256 | q->unplug_timer.data = (unsigned long)q; |
257 | 257 | ||
258 | /* | 258 | /* |
259 | * by default assume old behaviour and bounce for any highmem page | 259 | * by default assume old behaviour and bounce for any highmem page |
260 | */ | 260 | */ |
261 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); | 261 | blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); |
262 | 262 | ||
263 | blk_queue_activity_fn(q, NULL, NULL); | 263 | blk_queue_activity_fn(q, NULL, NULL); |
264 | } | 264 | } |
265 | 265 | ||
266 | EXPORT_SYMBOL(blk_queue_make_request); | 266 | EXPORT_SYMBOL(blk_queue_make_request); |
267 | 267 | ||
268 | static inline void rq_init(request_queue_t *q, struct request *rq) | 268 | static inline void rq_init(request_queue_t *q, struct request *rq) |
269 | { | 269 | { |
270 | INIT_LIST_HEAD(&rq->queuelist); | 270 | INIT_LIST_HEAD(&rq->queuelist); |
271 | 271 | ||
272 | rq->errors = 0; | 272 | rq->errors = 0; |
273 | rq->rq_status = RQ_ACTIVE; | 273 | rq->rq_status = RQ_ACTIVE; |
274 | rq->bio = rq->biotail = NULL; | 274 | rq->bio = rq->biotail = NULL; |
275 | rq->ioprio = 0; | 275 | rq->ioprio = 0; |
276 | rq->buffer = NULL; | 276 | rq->buffer = NULL; |
277 | rq->ref_count = 1; | 277 | rq->ref_count = 1; |
278 | rq->q = q; | 278 | rq->q = q; |
279 | rq->waiting = NULL; | 279 | rq->waiting = NULL; |
280 | rq->special = NULL; | 280 | rq->special = NULL; |
281 | rq->data_len = 0; | 281 | rq->data_len = 0; |
282 | rq->data = NULL; | 282 | rq->data = NULL; |
283 | rq->nr_phys_segments = 0; | 283 | rq->nr_phys_segments = 0; |
284 | rq->sense = NULL; | 284 | rq->sense = NULL; |
285 | rq->end_io = NULL; | 285 | rq->end_io = NULL; |
286 | rq->end_io_data = NULL; | 286 | rq->end_io_data = NULL; |
287 | } | 287 | } |
288 | 288 | ||
289 | /** | 289 | /** |
290 | * blk_queue_ordered - does this queue support ordered writes | 290 | * blk_queue_ordered - does this queue support ordered writes |
291 | * @q: the request queue | 291 | * @q: the request queue |
292 | * @flag: see below | 292 | * @flag: see below |
293 | * | 293 | * |
294 | * Description: | 294 | * Description: |
295 | * For journalled file systems, doing ordered writes on a commit | 295 | * For journalled file systems, doing ordered writes on a commit |
296 | * block instead of explicitly doing wait_on_buffer (which is bad | 296 | * block instead of explicitly doing wait_on_buffer (which is bad |
297 | * for performance) can be a big win. Block drivers supporting this | 297 | * for performance) can be a big win. Block drivers supporting this |
298 | * feature should call this function and indicate so. | 298 | * feature should call this function and indicate so. |
299 | * | 299 | * |
300 | **/ | 300 | **/ |
301 | void blk_queue_ordered(request_queue_t *q, int flag) | 301 | void blk_queue_ordered(request_queue_t *q, int flag) |
302 | { | 302 | { |
303 | switch (flag) { | 303 | switch (flag) { |
304 | case QUEUE_ORDERED_NONE: | 304 | case QUEUE_ORDERED_NONE: |
305 | if (q->flush_rq) | 305 | if (q->flush_rq) |
306 | kmem_cache_free(request_cachep, q->flush_rq); | 306 | kmem_cache_free(request_cachep, q->flush_rq); |
307 | q->flush_rq = NULL; | 307 | q->flush_rq = NULL; |
308 | q->ordered = flag; | 308 | q->ordered = flag; |
309 | break; | 309 | break; |
310 | case QUEUE_ORDERED_TAG: | 310 | case QUEUE_ORDERED_TAG: |
311 | q->ordered = flag; | 311 | q->ordered = flag; |
312 | break; | 312 | break; |
313 | case QUEUE_ORDERED_FLUSH: | 313 | case QUEUE_ORDERED_FLUSH: |
314 | q->ordered = flag; | 314 | q->ordered = flag; |
315 | if (!q->flush_rq) | 315 | if (!q->flush_rq) |
316 | q->flush_rq = kmem_cache_alloc(request_cachep, | 316 | q->flush_rq = kmem_cache_alloc(request_cachep, |
317 | GFP_KERNEL); | 317 | GFP_KERNEL); |
318 | break; | 318 | break; |
319 | default: | 319 | default: |
320 | printk("blk_queue_ordered: bad value %d\n", flag); | 320 | printk("blk_queue_ordered: bad value %d\n", flag); |
321 | break; | 321 | break; |
322 | } | 322 | } |
323 | } | 323 | } |
324 | 324 | ||
325 | EXPORT_SYMBOL(blk_queue_ordered); | 325 | EXPORT_SYMBOL(blk_queue_ordered); |
326 | 326 | ||
327 | /** | 327 | /** |
328 | * blk_queue_issue_flush_fn - set function for issuing a flush | 328 | * blk_queue_issue_flush_fn - set function for issuing a flush |
329 | * @q: the request queue | 329 | * @q: the request queue |
330 | * @iff: the function to be called issuing the flush | 330 | * @iff: the function to be called issuing the flush |
331 | * | 331 | * |
332 | * Description: | 332 | * Description: |
333 | * If a driver supports issuing a flush command, the support is notified | 333 | * If a driver supports issuing a flush command, the support is notified |
334 | * to the block layer by defining it through this call. | 334 | * to the block layer by defining it through this call. |
335 | * | 335 | * |
336 | **/ | 336 | **/ |
337 | void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff) | 337 | void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff) |
338 | { | 338 | { |
339 | q->issue_flush_fn = iff; | 339 | q->issue_flush_fn = iff; |
340 | } | 340 | } |
341 | 341 | ||
342 | EXPORT_SYMBOL(blk_queue_issue_flush_fn); | 342 | EXPORT_SYMBOL(blk_queue_issue_flush_fn); |
343 | 343 | ||
344 | /* | 344 | /* |
345 | * Cache flushing for ordered writes handling | 345 | * Cache flushing for ordered writes handling |
346 | */ | 346 | */ |
347 | static void blk_pre_flush_end_io(struct request *flush_rq) | 347 | static void blk_pre_flush_end_io(struct request *flush_rq) |
348 | { | 348 | { |
349 | struct request *rq = flush_rq->end_io_data; | 349 | struct request *rq = flush_rq->end_io_data; |
350 | request_queue_t *q = rq->q; | 350 | request_queue_t *q = rq->q; |
351 | 351 | ||
352 | elv_completed_request(q, flush_rq); | 352 | elv_completed_request(q, flush_rq); |
353 | 353 | ||
354 | rq->flags |= REQ_BAR_PREFLUSH; | 354 | rq->flags |= REQ_BAR_PREFLUSH; |
355 | 355 | ||
356 | if (!flush_rq->errors) | 356 | if (!flush_rq->errors) |
357 | elv_requeue_request(q, rq); | 357 | elv_requeue_request(q, rq); |
358 | else { | 358 | else { |
359 | q->end_flush_fn(q, flush_rq); | 359 | q->end_flush_fn(q, flush_rq); |
360 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | 360 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); |
361 | q->request_fn(q); | 361 | q->request_fn(q); |
362 | } | 362 | } |
363 | } | 363 | } |
364 | 364 | ||
365 | static void blk_post_flush_end_io(struct request *flush_rq) | 365 | static void blk_post_flush_end_io(struct request *flush_rq) |
366 | { | 366 | { |
367 | struct request *rq = flush_rq->end_io_data; | 367 | struct request *rq = flush_rq->end_io_data; |
368 | request_queue_t *q = rq->q; | 368 | request_queue_t *q = rq->q; |
369 | 369 | ||
370 | elv_completed_request(q, flush_rq); | 370 | elv_completed_request(q, flush_rq); |
371 | 371 | ||
372 | rq->flags |= REQ_BAR_POSTFLUSH; | 372 | rq->flags |= REQ_BAR_POSTFLUSH; |
373 | 373 | ||
374 | q->end_flush_fn(q, flush_rq); | 374 | q->end_flush_fn(q, flush_rq); |
375 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | 375 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); |
376 | q->request_fn(q); | 376 | q->request_fn(q); |
377 | } | 377 | } |
378 | 378 | ||
379 | struct request *blk_start_pre_flush(request_queue_t *q, struct request *rq) | 379 | struct request *blk_start_pre_flush(request_queue_t *q, struct request *rq) |
380 | { | 380 | { |
381 | struct request *flush_rq = q->flush_rq; | 381 | struct request *flush_rq = q->flush_rq; |
382 | 382 | ||
383 | BUG_ON(!blk_barrier_rq(rq)); | 383 | BUG_ON(!blk_barrier_rq(rq)); |
384 | 384 | ||
385 | if (test_and_set_bit(QUEUE_FLAG_FLUSH, &q->queue_flags)) | 385 | if (test_and_set_bit(QUEUE_FLAG_FLUSH, &q->queue_flags)) |
386 | return NULL; | 386 | return NULL; |
387 | 387 | ||
388 | rq_init(q, flush_rq); | 388 | rq_init(q, flush_rq); |
389 | flush_rq->elevator_private = NULL; | 389 | flush_rq->elevator_private = NULL; |
390 | flush_rq->flags = REQ_BAR_FLUSH; | 390 | flush_rq->flags = REQ_BAR_FLUSH; |
391 | flush_rq->rq_disk = rq->rq_disk; | 391 | flush_rq->rq_disk = rq->rq_disk; |
392 | flush_rq->rl = NULL; | 392 | flush_rq->rl = NULL; |
393 | 393 | ||
394 | /* | 394 | /* |
395 | * prepare_flush returns 0 if no flush is needed, just mark both | 395 | * prepare_flush returns 0 if no flush is needed, just mark both |
396 | * pre and post flush as done in that case | 396 | * pre and post flush as done in that case |
397 | */ | 397 | */ |
398 | if (!q->prepare_flush_fn(q, flush_rq)) { | 398 | if (!q->prepare_flush_fn(q, flush_rq)) { |
399 | rq->flags |= REQ_BAR_PREFLUSH | REQ_BAR_POSTFLUSH; | 399 | rq->flags |= REQ_BAR_PREFLUSH | REQ_BAR_POSTFLUSH; |
400 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | 400 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); |
401 | return rq; | 401 | return rq; |
402 | } | 402 | } |
403 | 403 | ||
404 | /* | 404 | /* |
405 | * some drivers dequeue requests right away, some only after io | 405 | * some drivers dequeue requests right away, some only after io |
406 | * completion. make sure the request is dequeued. | 406 | * completion. make sure the request is dequeued. |
407 | */ | 407 | */ |
408 | if (!list_empty(&rq->queuelist)) | 408 | if (!list_empty(&rq->queuelist)) |
409 | blkdev_dequeue_request(rq); | 409 | blkdev_dequeue_request(rq); |
410 | 410 | ||
411 | flush_rq->end_io_data = rq; | 411 | flush_rq->end_io_data = rq; |
412 | flush_rq->end_io = blk_pre_flush_end_io; | 412 | flush_rq->end_io = blk_pre_flush_end_io; |
413 | 413 | ||
414 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); | 414 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); |
415 | return flush_rq; | 415 | return flush_rq; |
416 | } | 416 | } |
417 | 417 | ||
418 | static void blk_start_post_flush(request_queue_t *q, struct request *rq) | 418 | static void blk_start_post_flush(request_queue_t *q, struct request *rq) |
419 | { | 419 | { |
420 | struct request *flush_rq = q->flush_rq; | 420 | struct request *flush_rq = q->flush_rq; |
421 | 421 | ||
422 | BUG_ON(!blk_barrier_rq(rq)); | 422 | BUG_ON(!blk_barrier_rq(rq)); |
423 | 423 | ||
424 | rq_init(q, flush_rq); | 424 | rq_init(q, flush_rq); |
425 | flush_rq->elevator_private = NULL; | 425 | flush_rq->elevator_private = NULL; |
426 | flush_rq->flags = REQ_BAR_FLUSH; | 426 | flush_rq->flags = REQ_BAR_FLUSH; |
427 | flush_rq->rq_disk = rq->rq_disk; | 427 | flush_rq->rq_disk = rq->rq_disk; |
428 | flush_rq->rl = NULL; | 428 | flush_rq->rl = NULL; |
429 | 429 | ||
430 | if (q->prepare_flush_fn(q, flush_rq)) { | 430 | if (q->prepare_flush_fn(q, flush_rq)) { |
431 | flush_rq->end_io_data = rq; | 431 | flush_rq->end_io_data = rq; |
432 | flush_rq->end_io = blk_post_flush_end_io; | 432 | flush_rq->end_io = blk_post_flush_end_io; |
433 | 433 | ||
434 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); | 434 | __elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0); |
435 | q->request_fn(q); | 435 | q->request_fn(q); |
436 | } | 436 | } |
437 | } | 437 | } |
438 | 438 | ||
439 | static inline int blk_check_end_barrier(request_queue_t *q, struct request *rq, | 439 | static inline int blk_check_end_barrier(request_queue_t *q, struct request *rq, |
440 | int sectors) | 440 | int sectors) |
441 | { | 441 | { |
442 | if (sectors > rq->nr_sectors) | 442 | if (sectors > rq->nr_sectors) |
443 | sectors = rq->nr_sectors; | 443 | sectors = rq->nr_sectors; |
444 | 444 | ||
445 | rq->nr_sectors -= sectors; | 445 | rq->nr_sectors -= sectors; |
446 | return rq->nr_sectors; | 446 | return rq->nr_sectors; |
447 | } | 447 | } |
448 | 448 | ||
449 | static int __blk_complete_barrier_rq(request_queue_t *q, struct request *rq, | 449 | static int __blk_complete_barrier_rq(request_queue_t *q, struct request *rq, |
450 | int sectors, int queue_locked) | 450 | int sectors, int queue_locked) |
451 | { | 451 | { |
452 | if (q->ordered != QUEUE_ORDERED_FLUSH) | 452 | if (q->ordered != QUEUE_ORDERED_FLUSH) |
453 | return 0; | 453 | return 0; |
454 | if (!blk_fs_request(rq) || !blk_barrier_rq(rq)) | 454 | if (!blk_fs_request(rq) || !blk_barrier_rq(rq)) |
455 | return 0; | 455 | return 0; |
456 | if (blk_barrier_postflush(rq)) | 456 | if (blk_barrier_postflush(rq)) |
457 | return 0; | 457 | return 0; |
458 | 458 | ||
459 | if (!blk_check_end_barrier(q, rq, sectors)) { | 459 | if (!blk_check_end_barrier(q, rq, sectors)) { |
460 | unsigned long flags = 0; | 460 | unsigned long flags = 0; |
461 | 461 | ||
462 | if (!queue_locked) | 462 | if (!queue_locked) |
463 | spin_lock_irqsave(q->queue_lock, flags); | 463 | spin_lock_irqsave(q->queue_lock, flags); |
464 | 464 | ||
465 | blk_start_post_flush(q, rq); | 465 | blk_start_post_flush(q, rq); |
466 | 466 | ||
467 | if (!queue_locked) | 467 | if (!queue_locked) |
468 | spin_unlock_irqrestore(q->queue_lock, flags); | 468 | spin_unlock_irqrestore(q->queue_lock, flags); |
469 | } | 469 | } |
470 | 470 | ||
471 | return 1; | 471 | return 1; |
472 | } | 472 | } |
473 | 473 | ||
474 | /** | 474 | /** |
475 | * blk_complete_barrier_rq - complete possible barrier request | 475 | * blk_complete_barrier_rq - complete possible barrier request |
476 | * @q: the request queue for the device | 476 | * @q: the request queue for the device |
477 | * @rq: the request | 477 | * @rq: the request |
478 | * @sectors: number of sectors to complete | 478 | * @sectors: number of sectors to complete |
479 | * | 479 | * |
480 | * Description: | 480 | * Description: |
481 | * Used in driver end_io handling to determine whether to postpone | 481 | * Used in driver end_io handling to determine whether to postpone |
482 | * completion of a barrier request until a post flush has been done. This | 482 | * completion of a barrier request until a post flush has been done. This |
483 | * is the unlocked variant, used if the caller doesn't already hold the | 483 | * is the unlocked variant, used if the caller doesn't already hold the |
484 | * queue lock. | 484 | * queue lock. |
485 | **/ | 485 | **/ |
486 | int blk_complete_barrier_rq(request_queue_t *q, struct request *rq, int sectors) | 486 | int blk_complete_barrier_rq(request_queue_t *q, struct request *rq, int sectors) |
487 | { | 487 | { |
488 | return __blk_complete_barrier_rq(q, rq, sectors, 0); | 488 | return __blk_complete_barrier_rq(q, rq, sectors, 0); |
489 | } | 489 | } |
490 | EXPORT_SYMBOL(blk_complete_barrier_rq); | 490 | EXPORT_SYMBOL(blk_complete_barrier_rq); |
491 | 491 | ||
492 | /** | 492 | /** |
493 | * blk_complete_barrier_rq_locked - complete possible barrier request | 493 | * blk_complete_barrier_rq_locked - complete possible barrier request |
494 | * @q: the request queue for the device | 494 | * @q: the request queue for the device |
495 | * @rq: the request | 495 | * @rq: the request |
496 | * @sectors: number of sectors to complete | 496 | * @sectors: number of sectors to complete |
497 | * | 497 | * |
498 | * Description: | 498 | * Description: |
499 | * See blk_complete_barrier_rq(). This variant must be used if the caller | 499 | * See blk_complete_barrier_rq(). This variant must be used if the caller |
500 | * holds the queue lock. | 500 | * holds the queue lock. |
501 | **/ | 501 | **/ |
502 | int blk_complete_barrier_rq_locked(request_queue_t *q, struct request *rq, | 502 | int blk_complete_barrier_rq_locked(request_queue_t *q, struct request *rq, |
503 | int sectors) | 503 | int sectors) |
504 | { | 504 | { |
505 | return __blk_complete_barrier_rq(q, rq, sectors, 1); | 505 | return __blk_complete_barrier_rq(q, rq, sectors, 1); |
506 | } | 506 | } |
507 | EXPORT_SYMBOL(blk_complete_barrier_rq_locked); | 507 | EXPORT_SYMBOL(blk_complete_barrier_rq_locked); |
508 | 508 | ||
509 | /** | 509 | /** |
510 | * blk_queue_bounce_limit - set bounce buffer limit for queue | 510 | * blk_queue_bounce_limit - set bounce buffer limit for queue |
511 | * @q: the request queue for the device | 511 | * @q: the request queue for the device |
512 | * @dma_addr: bus address limit | 512 | * @dma_addr: bus address limit |
513 | * | 513 | * |
514 | * Description: | 514 | * Description: |
515 | * Different hardware can have different requirements as to what pages | 515 | * Different hardware can have different requirements as to what pages |
516 | * it can do I/O directly to. A low level driver can call | 516 | * it can do I/O directly to. A low level driver can call |
517 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce | 517 | * blk_queue_bounce_limit to have lower memory pages allocated as bounce |
518 | * buffers for doing I/O to pages residing above @page. By default | 518 | * buffers for doing I/O to pages residing above @page. By default |
519 | * the block layer sets this to the highest numbered "low" memory page. | 519 | * the block layer sets this to the highest numbered "low" memory page. |
520 | **/ | 520 | **/ |
521 | void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) | 521 | void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) |
522 | { | 522 | { |
523 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; | 523 | unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; |
524 | 524 | ||
525 | /* | 525 | /* |
526 | * set appropriate bounce gfp mask -- unfortunately we don't have a | 526 | * set appropriate bounce gfp mask -- unfortunately we don't have a |
527 | * full 4GB zone, so we have to resort to low memory for any bounces. | 527 | * full 4GB zone, so we have to resort to low memory for any bounces. |
528 | * ISA has its own < 16MB zone. | 528 | * ISA has its own < 16MB zone. |
529 | */ | 529 | */ |
530 | if (bounce_pfn < blk_max_low_pfn) { | 530 | if (bounce_pfn < blk_max_low_pfn) { |
531 | BUG_ON(dma_addr < BLK_BOUNCE_ISA); | 531 | BUG_ON(dma_addr < BLK_BOUNCE_ISA); |
532 | init_emergency_isa_pool(); | 532 | init_emergency_isa_pool(); |
533 | q->bounce_gfp = GFP_NOIO | GFP_DMA; | 533 | q->bounce_gfp = GFP_NOIO | GFP_DMA; |
534 | } else | 534 | } else |
535 | q->bounce_gfp = GFP_NOIO; | 535 | q->bounce_gfp = GFP_NOIO; |
536 | 536 | ||
537 | q->bounce_pfn = bounce_pfn; | 537 | q->bounce_pfn = bounce_pfn; |
538 | } | 538 | } |
539 | 539 | ||
540 | EXPORT_SYMBOL(blk_queue_bounce_limit); | 540 | EXPORT_SYMBOL(blk_queue_bounce_limit); |
541 | 541 | ||
542 | /** | 542 | /** |
543 | * blk_queue_max_sectors - set max sectors for a request for this queue | 543 | * blk_queue_max_sectors - set max sectors for a request for this queue |
544 | * @q: the request queue for the device | 544 | * @q: the request queue for the device |
545 | * @max_sectors: max sectors in the usual 512b unit | 545 | * @max_sectors: max sectors in the usual 512b unit |
546 | * | 546 | * |
547 | * Description: | 547 | * Description: |
548 | * Enables a low level driver to set an upper limit on the size of | 548 | * Enables a low level driver to set an upper limit on the size of |
549 | * received requests. | 549 | * received requests. |
550 | **/ | 550 | **/ |
551 | void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors) | 551 | void blk_queue_max_sectors(request_queue_t *q, unsigned short max_sectors) |
552 | { | 552 | { |
553 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { | 553 | if ((max_sectors << 9) < PAGE_CACHE_SIZE) { |
554 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); | 554 | max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); |
555 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); | 555 | printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); |
556 | } | 556 | } |
557 | 557 | ||
558 | if (BLK_DEF_MAX_SECTORS > max_sectors) | 558 | if (BLK_DEF_MAX_SECTORS > max_sectors) |
559 | q->max_hw_sectors = q->max_sectors = max_sectors; | 559 | q->max_hw_sectors = q->max_sectors = max_sectors; |
560 | else { | 560 | else { |
561 | q->max_sectors = BLK_DEF_MAX_SECTORS; | 561 | q->max_sectors = BLK_DEF_MAX_SECTORS; |
562 | q->max_hw_sectors = max_sectors; | 562 | q->max_hw_sectors = max_sectors; |
563 | } | 563 | } |
564 | } | 564 | } |
565 | 565 | ||
566 | EXPORT_SYMBOL(blk_queue_max_sectors); | 566 | EXPORT_SYMBOL(blk_queue_max_sectors); |
567 | 567 | ||
568 | /** | 568 | /** |
569 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue | 569 | * blk_queue_max_phys_segments - set max phys segments for a request for this queue |
570 | * @q: the request queue for the device | 570 | * @q: the request queue for the device |
571 | * @max_segments: max number of segments | 571 | * @max_segments: max number of segments |
572 | * | 572 | * |
573 | * Description: | 573 | * Description: |
574 | * Enables a low level driver to set an upper limit on the number of | 574 | * Enables a low level driver to set an upper limit on the number of |
575 | * physical data segments in a request. This would be the largest sized | 575 | * physical data segments in a request. This would be the largest sized |
576 | * scatter list the driver could handle. | 576 | * scatter list the driver could handle. |
577 | **/ | 577 | **/ |
578 | void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments) | 578 | void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments) |
579 | { | 579 | { |
580 | if (!max_segments) { | 580 | if (!max_segments) { |
581 | max_segments = 1; | 581 | max_segments = 1; |
582 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | 582 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); |
583 | } | 583 | } |
584 | 584 | ||
585 | q->max_phys_segments = max_segments; | 585 | q->max_phys_segments = max_segments; |
586 | } | 586 | } |
587 | 587 | ||
588 | EXPORT_SYMBOL(blk_queue_max_phys_segments); | 588 | EXPORT_SYMBOL(blk_queue_max_phys_segments); |
589 | 589 | ||
590 | /** | 590 | /** |
591 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue | 591 | * blk_queue_max_hw_segments - set max hw segments for a request for this queue |
592 | * @q: the request queue for the device | 592 | * @q: the request queue for the device |
593 | * @max_segments: max number of segments | 593 | * @max_segments: max number of segments |
594 | * | 594 | * |
595 | * Description: | 595 | * Description: |
596 | * Enables a low level driver to set an upper limit on the number of | 596 | * Enables a low level driver to set an upper limit on the number of |
597 | * hw data segments in a request. This would be the largest number of | 597 | * hw data segments in a request. This would be the largest number of |
598 | * address/length pairs the host adapter can actually give as once | 598 | * address/length pairs the host adapter can actually give as once |
599 | * to the device. | 599 | * to the device. |
600 | **/ | 600 | **/ |
601 | void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments) | 601 | void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments) |
602 | { | 602 | { |
603 | if (!max_segments) { | 603 | if (!max_segments) { |
604 | max_segments = 1; | 604 | max_segments = 1; |
605 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); | 605 | printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); |
606 | } | 606 | } |
607 | 607 | ||
608 | q->max_hw_segments = max_segments; | 608 | q->max_hw_segments = max_segments; |
609 | } | 609 | } |
610 | 610 | ||
611 | EXPORT_SYMBOL(blk_queue_max_hw_segments); | 611 | EXPORT_SYMBOL(blk_queue_max_hw_segments); |
612 | 612 | ||
613 | /** | 613 | /** |
614 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg | 614 | * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg |
615 | * @q: the request queue for the device | 615 | * @q: the request queue for the device |
616 | * @max_size: max size of segment in bytes | 616 | * @max_size: max size of segment in bytes |
617 | * | 617 | * |
618 | * Description: | 618 | * Description: |
619 | * Enables a low level driver to set an upper limit on the size of a | 619 | * Enables a low level driver to set an upper limit on the size of a |
620 | * coalesced segment | 620 | * coalesced segment |
621 | **/ | 621 | **/ |
622 | void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size) | 622 | void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size) |
623 | { | 623 | { |
624 | if (max_size < PAGE_CACHE_SIZE) { | 624 | if (max_size < PAGE_CACHE_SIZE) { |
625 | max_size = PAGE_CACHE_SIZE; | 625 | max_size = PAGE_CACHE_SIZE; |
626 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); | 626 | printk("%s: set to minimum %d\n", __FUNCTION__, max_size); |
627 | } | 627 | } |
628 | 628 | ||
629 | q->max_segment_size = max_size; | 629 | q->max_segment_size = max_size; |
630 | } | 630 | } |
631 | 631 | ||
632 | EXPORT_SYMBOL(blk_queue_max_segment_size); | 632 | EXPORT_SYMBOL(blk_queue_max_segment_size); |
633 | 633 | ||
634 | /** | 634 | /** |
635 | * blk_queue_hardsect_size - set hardware sector size for the queue | 635 | * blk_queue_hardsect_size - set hardware sector size for the queue |
636 | * @q: the request queue for the device | 636 | * @q: the request queue for the device |
637 | * @size: the hardware sector size, in bytes | 637 | * @size: the hardware sector size, in bytes |
638 | * | 638 | * |
639 | * Description: | 639 | * Description: |
640 | * This should typically be set to the lowest possible sector size | 640 | * This should typically be set to the lowest possible sector size |
641 | * that the hardware can operate on (possible without reverting to | 641 | * that the hardware can operate on (possible without reverting to |
642 | * even internal read-modify-write operations). Usually the default | 642 | * even internal read-modify-write operations). Usually the default |
643 | * of 512 covers most hardware. | 643 | * of 512 covers most hardware. |
644 | **/ | 644 | **/ |
645 | void blk_queue_hardsect_size(request_queue_t *q, unsigned short size) | 645 | void blk_queue_hardsect_size(request_queue_t *q, unsigned short size) |
646 | { | 646 | { |
647 | q->hardsect_size = size; | 647 | q->hardsect_size = size; |
648 | } | 648 | } |
649 | 649 | ||
650 | EXPORT_SYMBOL(blk_queue_hardsect_size); | 650 | EXPORT_SYMBOL(blk_queue_hardsect_size); |
651 | 651 | ||
652 | /* | 652 | /* |
653 | * Returns the minimum that is _not_ zero, unless both are zero. | 653 | * Returns the minimum that is _not_ zero, unless both are zero. |
654 | */ | 654 | */ |
655 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) | 655 | #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) |
656 | 656 | ||
657 | /** | 657 | /** |
658 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers | 658 | * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers |
659 | * @t: the stacking driver (top) | 659 | * @t: the stacking driver (top) |
660 | * @b: the underlying device (bottom) | 660 | * @b: the underlying device (bottom) |
661 | **/ | 661 | **/ |
662 | void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b) | 662 | void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b) |
663 | { | 663 | { |
664 | /* zero is "infinity" */ | 664 | /* zero is "infinity" */ |
665 | t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); | 665 | t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); |
666 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); | 666 | t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); |
667 | 667 | ||
668 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); | 668 | t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); |
669 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); | 669 | t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); |
670 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); | 670 | t->max_segment_size = min(t->max_segment_size,b->max_segment_size); |
671 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); | 671 | t->hardsect_size = max(t->hardsect_size,b->hardsect_size); |
672 | } | 672 | } |
673 | 673 | ||
674 | EXPORT_SYMBOL(blk_queue_stack_limits); | 674 | EXPORT_SYMBOL(blk_queue_stack_limits); |
675 | 675 | ||
676 | /** | 676 | /** |
677 | * blk_queue_segment_boundary - set boundary rules for segment merging | 677 | * blk_queue_segment_boundary - set boundary rules for segment merging |
678 | * @q: the request queue for the device | 678 | * @q: the request queue for the device |
679 | * @mask: the memory boundary mask | 679 | * @mask: the memory boundary mask |
680 | **/ | 680 | **/ |
681 | void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask) | 681 | void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask) |
682 | { | 682 | { |
683 | if (mask < PAGE_CACHE_SIZE - 1) { | 683 | if (mask < PAGE_CACHE_SIZE - 1) { |
684 | mask = PAGE_CACHE_SIZE - 1; | 684 | mask = PAGE_CACHE_SIZE - 1; |
685 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); | 685 | printk("%s: set to minimum %lx\n", __FUNCTION__, mask); |
686 | } | 686 | } |
687 | 687 | ||
688 | q->seg_boundary_mask = mask; | 688 | q->seg_boundary_mask = mask; |
689 | } | 689 | } |
690 | 690 | ||
691 | EXPORT_SYMBOL(blk_queue_segment_boundary); | 691 | EXPORT_SYMBOL(blk_queue_segment_boundary); |
692 | 692 | ||
693 | /** | 693 | /** |
694 | * blk_queue_dma_alignment - set dma length and memory alignment | 694 | * blk_queue_dma_alignment - set dma length and memory alignment |
695 | * @q: the request queue for the device | 695 | * @q: the request queue for the device |
696 | * @mask: alignment mask | 696 | * @mask: alignment mask |
697 | * | 697 | * |
698 | * description: | 698 | * description: |
699 | * set required memory and length aligment for direct dma transactions. | 699 | * set required memory and length aligment for direct dma transactions. |
700 | * this is used when buiding direct io requests for the queue. | 700 | * this is used when buiding direct io requests for the queue. |
701 | * | 701 | * |
702 | **/ | 702 | **/ |
703 | void blk_queue_dma_alignment(request_queue_t *q, int mask) | 703 | void blk_queue_dma_alignment(request_queue_t *q, int mask) |
704 | { | 704 | { |
705 | q->dma_alignment = mask; | 705 | q->dma_alignment = mask; |
706 | } | 706 | } |
707 | 707 | ||
708 | EXPORT_SYMBOL(blk_queue_dma_alignment); | 708 | EXPORT_SYMBOL(blk_queue_dma_alignment); |
709 | 709 | ||
710 | /** | 710 | /** |
711 | * blk_queue_find_tag - find a request by its tag and queue | 711 | * blk_queue_find_tag - find a request by its tag and queue |
712 | * @q: The request queue for the device | 712 | * @q: The request queue for the device |
713 | * @tag: The tag of the request | 713 | * @tag: The tag of the request |
714 | * | 714 | * |
715 | * Notes: | 715 | * Notes: |
716 | * Should be used when a device returns a tag and you want to match | 716 | * Should be used when a device returns a tag and you want to match |
717 | * it with a request. | 717 | * it with a request. |
718 | * | 718 | * |
719 | * no locks need be held. | 719 | * no locks need be held. |
720 | **/ | 720 | **/ |
721 | struct request *blk_queue_find_tag(request_queue_t *q, int tag) | 721 | struct request *blk_queue_find_tag(request_queue_t *q, int tag) |
722 | { | 722 | { |
723 | struct blk_queue_tag *bqt = q->queue_tags; | 723 | struct blk_queue_tag *bqt = q->queue_tags; |
724 | 724 | ||
725 | if (unlikely(bqt == NULL || tag >= bqt->real_max_depth)) | 725 | if (unlikely(bqt == NULL || tag >= bqt->real_max_depth)) |
726 | return NULL; | 726 | return NULL; |
727 | 727 | ||
728 | return bqt->tag_index[tag]; | 728 | return bqt->tag_index[tag]; |
729 | } | 729 | } |
730 | 730 | ||
731 | EXPORT_SYMBOL(blk_queue_find_tag); | 731 | EXPORT_SYMBOL(blk_queue_find_tag); |
732 | 732 | ||
733 | /** | 733 | /** |
734 | * __blk_queue_free_tags - release tag maintenance info | 734 | * __blk_queue_free_tags - release tag maintenance info |
735 | * @q: the request queue for the device | 735 | * @q: the request queue for the device |
736 | * | 736 | * |
737 | * Notes: | 737 | * Notes: |
738 | * blk_cleanup_queue() will take care of calling this function, if tagging | 738 | * blk_cleanup_queue() will take care of calling this function, if tagging |
739 | * has been used. So there's no need to call this directly. | 739 | * has been used. So there's no need to call this directly. |
740 | **/ | 740 | **/ |
741 | static void __blk_queue_free_tags(request_queue_t *q) | 741 | static void __blk_queue_free_tags(request_queue_t *q) |
742 | { | 742 | { |
743 | struct blk_queue_tag *bqt = q->queue_tags; | 743 | struct blk_queue_tag *bqt = q->queue_tags; |
744 | 744 | ||
745 | if (!bqt) | 745 | if (!bqt) |
746 | return; | 746 | return; |
747 | 747 | ||
748 | if (atomic_dec_and_test(&bqt->refcnt)) { | 748 | if (atomic_dec_and_test(&bqt->refcnt)) { |
749 | BUG_ON(bqt->busy); | 749 | BUG_ON(bqt->busy); |
750 | BUG_ON(!list_empty(&bqt->busy_list)); | 750 | BUG_ON(!list_empty(&bqt->busy_list)); |
751 | 751 | ||
752 | kfree(bqt->tag_index); | 752 | kfree(bqt->tag_index); |
753 | bqt->tag_index = NULL; | 753 | bqt->tag_index = NULL; |
754 | 754 | ||
755 | kfree(bqt->tag_map); | 755 | kfree(bqt->tag_map); |
756 | bqt->tag_map = NULL; | 756 | bqt->tag_map = NULL; |
757 | 757 | ||
758 | kfree(bqt); | 758 | kfree(bqt); |
759 | } | 759 | } |
760 | 760 | ||
761 | q->queue_tags = NULL; | 761 | q->queue_tags = NULL; |
762 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); | 762 | q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); |
763 | } | 763 | } |
764 | 764 | ||
765 | /** | 765 | /** |
766 | * blk_queue_free_tags - release tag maintenance info | 766 | * blk_queue_free_tags - release tag maintenance info |
767 | * @q: the request queue for the device | 767 | * @q: the request queue for the device |
768 | * | 768 | * |
769 | * Notes: | 769 | * Notes: |
770 | * This is used to disabled tagged queuing to a device, yet leave | 770 | * This is used to disabled tagged queuing to a device, yet leave |
771 | * queue in function. | 771 | * queue in function. |
772 | **/ | 772 | **/ |
773 | void blk_queue_free_tags(request_queue_t *q) | 773 | void blk_queue_free_tags(request_queue_t *q) |
774 | { | 774 | { |
775 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | 775 | clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); |
776 | } | 776 | } |
777 | 777 | ||
778 | EXPORT_SYMBOL(blk_queue_free_tags); | 778 | EXPORT_SYMBOL(blk_queue_free_tags); |
779 | 779 | ||
780 | static int | 780 | static int |
781 | init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) | 781 | init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) |
782 | { | 782 | { |
783 | struct request **tag_index; | 783 | struct request **tag_index; |
784 | unsigned long *tag_map; | 784 | unsigned long *tag_map; |
785 | int nr_ulongs; | 785 | int nr_ulongs; |
786 | 786 | ||
787 | if (depth > q->nr_requests * 2) { | 787 | if (depth > q->nr_requests * 2) { |
788 | depth = q->nr_requests * 2; | 788 | depth = q->nr_requests * 2; |
789 | printk(KERN_ERR "%s: adjusted depth to %d\n", | 789 | printk(KERN_ERR "%s: adjusted depth to %d\n", |
790 | __FUNCTION__, depth); | 790 | __FUNCTION__, depth); |
791 | } | 791 | } |
792 | 792 | ||
793 | tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC); | 793 | tag_index = kmalloc(depth * sizeof(struct request *), GFP_ATOMIC); |
794 | if (!tag_index) | 794 | if (!tag_index) |
795 | goto fail; | 795 | goto fail; |
796 | 796 | ||
797 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; | 797 | nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; |
798 | tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); | 798 | tag_map = kmalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); |
799 | if (!tag_map) | 799 | if (!tag_map) |
800 | goto fail; | 800 | goto fail; |
801 | 801 | ||
802 | memset(tag_index, 0, depth * sizeof(struct request *)); | 802 | memset(tag_index, 0, depth * sizeof(struct request *)); |
803 | memset(tag_map, 0, nr_ulongs * sizeof(unsigned long)); | 803 | memset(tag_map, 0, nr_ulongs * sizeof(unsigned long)); |
804 | tags->real_max_depth = depth; | 804 | tags->real_max_depth = depth; |
805 | tags->max_depth = depth; | 805 | tags->max_depth = depth; |
806 | tags->tag_index = tag_index; | 806 | tags->tag_index = tag_index; |
807 | tags->tag_map = tag_map; | 807 | tags->tag_map = tag_map; |
808 | 808 | ||
809 | return 0; | 809 | return 0; |
810 | fail: | 810 | fail: |
811 | kfree(tag_index); | 811 | kfree(tag_index); |
812 | return -ENOMEM; | 812 | return -ENOMEM; |
813 | } | 813 | } |
814 | 814 | ||
815 | /** | 815 | /** |
816 | * blk_queue_init_tags - initialize the queue tag info | 816 | * blk_queue_init_tags - initialize the queue tag info |
817 | * @q: the request queue for the device | 817 | * @q: the request queue for the device |
818 | * @depth: the maximum queue depth supported | 818 | * @depth: the maximum queue depth supported |
819 | * @tags: the tag to use | 819 | * @tags: the tag to use |
820 | **/ | 820 | **/ |
821 | int blk_queue_init_tags(request_queue_t *q, int depth, | 821 | int blk_queue_init_tags(request_queue_t *q, int depth, |
822 | struct blk_queue_tag *tags) | 822 | struct blk_queue_tag *tags) |
823 | { | 823 | { |
824 | int rc; | 824 | int rc; |
825 | 825 | ||
826 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); | 826 | BUG_ON(tags && q->queue_tags && tags != q->queue_tags); |
827 | 827 | ||
828 | if (!tags && !q->queue_tags) { | 828 | if (!tags && !q->queue_tags) { |
829 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); | 829 | tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); |
830 | if (!tags) | 830 | if (!tags) |
831 | goto fail; | 831 | goto fail; |
832 | 832 | ||
833 | if (init_tag_map(q, tags, depth)) | 833 | if (init_tag_map(q, tags, depth)) |
834 | goto fail; | 834 | goto fail; |
835 | 835 | ||
836 | INIT_LIST_HEAD(&tags->busy_list); | 836 | INIT_LIST_HEAD(&tags->busy_list); |
837 | tags->busy = 0; | 837 | tags->busy = 0; |
838 | atomic_set(&tags->refcnt, 1); | 838 | atomic_set(&tags->refcnt, 1); |
839 | } else if (q->queue_tags) { | 839 | } else if (q->queue_tags) { |
840 | if ((rc = blk_queue_resize_tags(q, depth))) | 840 | if ((rc = blk_queue_resize_tags(q, depth))) |
841 | return rc; | 841 | return rc; |
842 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); | 842 | set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); |
843 | return 0; | 843 | return 0; |
844 | } else | 844 | } else |
845 | atomic_inc(&tags->refcnt); | 845 | atomic_inc(&tags->refcnt); |
846 | 846 | ||
847 | /* | 847 | /* |
848 | * assign it, all done | 848 | * assign it, all done |
849 | */ | 849 | */ |
850 | q->queue_tags = tags; | 850 | q->queue_tags = tags; |
851 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); | 851 | q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); |
852 | return 0; | 852 | return 0; |
853 | fail: | 853 | fail: |
854 | kfree(tags); | 854 | kfree(tags); |
855 | return -ENOMEM; | 855 | return -ENOMEM; |
856 | } | 856 | } |
857 | 857 | ||
858 | EXPORT_SYMBOL(blk_queue_init_tags); | 858 | EXPORT_SYMBOL(blk_queue_init_tags); |
859 | 859 | ||
860 | /** | 860 | /** |
861 | * blk_queue_resize_tags - change the queueing depth | 861 | * blk_queue_resize_tags - change the queueing depth |
862 | * @q: the request queue for the device | 862 | * @q: the request queue for the device |
863 | * @new_depth: the new max command queueing depth | 863 | * @new_depth: the new max command queueing depth |
864 | * | 864 | * |
865 | * Notes: | 865 | * Notes: |
866 | * Must be called with the queue lock held. | 866 | * Must be called with the queue lock held. |
867 | **/ | 867 | **/ |
868 | int blk_queue_resize_tags(request_queue_t *q, int new_depth) | 868 | int blk_queue_resize_tags(request_queue_t *q, int new_depth) |
869 | { | 869 | { |
870 | struct blk_queue_tag *bqt = q->queue_tags; | 870 | struct blk_queue_tag *bqt = q->queue_tags; |
871 | struct request **tag_index; | 871 | struct request **tag_index; |
872 | unsigned long *tag_map; | 872 | unsigned long *tag_map; |
873 | int max_depth, nr_ulongs; | 873 | int max_depth, nr_ulongs; |
874 | 874 | ||
875 | if (!bqt) | 875 | if (!bqt) |
876 | return -ENXIO; | 876 | return -ENXIO; |
877 | 877 | ||
878 | /* | 878 | /* |
879 | * if we already have large enough real_max_depth. just | 879 | * if we already have large enough real_max_depth. just |
880 | * adjust max_depth. *NOTE* as requests with tag value | 880 | * adjust max_depth. *NOTE* as requests with tag value |
881 | * between new_depth and real_max_depth can be in-flight, tag | 881 | * between new_depth and real_max_depth can be in-flight, tag |
882 | * map can not be shrunk blindly here. | 882 | * map can not be shrunk blindly here. |
883 | */ | 883 | */ |
884 | if (new_depth <= bqt->real_max_depth) { | 884 | if (new_depth <= bqt->real_max_depth) { |
885 | bqt->max_depth = new_depth; | 885 | bqt->max_depth = new_depth; |
886 | return 0; | 886 | return 0; |
887 | } | 887 | } |
888 | 888 | ||
889 | /* | 889 | /* |
890 | * save the old state info, so we can copy it back | 890 | * save the old state info, so we can copy it back |
891 | */ | 891 | */ |
892 | tag_index = bqt->tag_index; | 892 | tag_index = bqt->tag_index; |
893 | tag_map = bqt->tag_map; | 893 | tag_map = bqt->tag_map; |
894 | max_depth = bqt->real_max_depth; | 894 | max_depth = bqt->real_max_depth; |
895 | 895 | ||
896 | if (init_tag_map(q, bqt, new_depth)) | 896 | if (init_tag_map(q, bqt, new_depth)) |
897 | return -ENOMEM; | 897 | return -ENOMEM; |
898 | 898 | ||
899 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); | 899 | memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); |
900 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; | 900 | nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; |
901 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); | 901 | memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); |
902 | 902 | ||
903 | kfree(tag_index); | 903 | kfree(tag_index); |
904 | kfree(tag_map); | 904 | kfree(tag_map); |
905 | return 0; | 905 | return 0; |
906 | } | 906 | } |
907 | 907 | ||
908 | EXPORT_SYMBOL(blk_queue_resize_tags); | 908 | EXPORT_SYMBOL(blk_queue_resize_tags); |
909 | 909 | ||
910 | /** | 910 | /** |
911 | * blk_queue_end_tag - end tag operations for a request | 911 | * blk_queue_end_tag - end tag operations for a request |
912 | * @q: the request queue for the device | 912 | * @q: the request queue for the device |
913 | * @rq: the request that has completed | 913 | * @rq: the request that has completed |
914 | * | 914 | * |
915 | * Description: | 915 | * Description: |
916 | * Typically called when end_that_request_first() returns 0, meaning | 916 | * Typically called when end_that_request_first() returns 0, meaning |
917 | * all transfers have been done for a request. It's important to call | 917 | * all transfers have been done for a request. It's important to call |
918 | * this function before end_that_request_last(), as that will put the | 918 | * this function before end_that_request_last(), as that will put the |
919 | * request back on the free list thus corrupting the internal tag list. | 919 | * request back on the free list thus corrupting the internal tag list. |
920 | * | 920 | * |
921 | * Notes: | 921 | * Notes: |
922 | * queue lock must be held. | 922 | * queue lock must be held. |
923 | **/ | 923 | **/ |
924 | void blk_queue_end_tag(request_queue_t *q, struct request *rq) | 924 | void blk_queue_end_tag(request_queue_t *q, struct request *rq) |
925 | { | 925 | { |
926 | struct blk_queue_tag *bqt = q->queue_tags; | 926 | struct blk_queue_tag *bqt = q->queue_tags; |
927 | int tag = rq->tag; | 927 | int tag = rq->tag; |
928 | 928 | ||
929 | BUG_ON(tag == -1); | 929 | BUG_ON(tag == -1); |
930 | 930 | ||
931 | if (unlikely(tag >= bqt->real_max_depth)) | 931 | if (unlikely(tag >= bqt->real_max_depth)) |
932 | /* | 932 | /* |
933 | * This can happen after tag depth has been reduced. | 933 | * This can happen after tag depth has been reduced. |
934 | * FIXME: how about a warning or info message here? | 934 | * FIXME: how about a warning or info message here? |
935 | */ | 935 | */ |
936 | return; | 936 | return; |
937 | 937 | ||
938 | if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { | 938 | if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { |
939 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", | 939 | printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", |
940 | __FUNCTION__, tag); | 940 | __FUNCTION__, tag); |
941 | return; | 941 | return; |
942 | } | 942 | } |
943 | 943 | ||
944 | list_del_init(&rq->queuelist); | 944 | list_del_init(&rq->queuelist); |
945 | rq->flags &= ~REQ_QUEUED; | 945 | rq->flags &= ~REQ_QUEUED; |
946 | rq->tag = -1; | 946 | rq->tag = -1; |
947 | 947 | ||
948 | if (unlikely(bqt->tag_index[tag] == NULL)) | 948 | if (unlikely(bqt->tag_index[tag] == NULL)) |
949 | printk(KERN_ERR "%s: tag %d is missing\n", | 949 | printk(KERN_ERR "%s: tag %d is missing\n", |
950 | __FUNCTION__, tag); | 950 | __FUNCTION__, tag); |
951 | 951 | ||
952 | bqt->tag_index[tag] = NULL; | 952 | bqt->tag_index[tag] = NULL; |
953 | bqt->busy--; | 953 | bqt->busy--; |
954 | } | 954 | } |
955 | 955 | ||
956 | EXPORT_SYMBOL(blk_queue_end_tag); | 956 | EXPORT_SYMBOL(blk_queue_end_tag); |
957 | 957 | ||
958 | /** | 958 | /** |
959 | * blk_queue_start_tag - find a free tag and assign it | 959 | * blk_queue_start_tag - find a free tag and assign it |
960 | * @q: the request queue for the device | 960 | * @q: the request queue for the device |
961 | * @rq: the block request that needs tagging | 961 | * @rq: the block request that needs tagging |
962 | * | 962 | * |
963 | * Description: | 963 | * Description: |
964 | * This can either be used as a stand-alone helper, or possibly be | 964 | * This can either be used as a stand-alone helper, or possibly be |
965 | * assigned as the queue &prep_rq_fn (in which case &struct request | 965 | * assigned as the queue &prep_rq_fn (in which case &struct request |
966 | * automagically gets a tag assigned). Note that this function | 966 | * automagically gets a tag assigned). Note that this function |
967 | * assumes that any type of request can be queued! if this is not | 967 | * assumes that any type of request can be queued! if this is not |
968 | * true for your device, you must check the request type before | 968 | * true for your device, you must check the request type before |
969 | * calling this function. The request will also be removed from | 969 | * calling this function. The request will also be removed from |
970 | * the request queue, so it's the drivers responsibility to readd | 970 | * the request queue, so it's the drivers responsibility to readd |
971 | * it if it should need to be restarted for some reason. | 971 | * it if it should need to be restarted for some reason. |
972 | * | 972 | * |
973 | * Notes: | 973 | * Notes: |
974 | * queue lock must be held. | 974 | * queue lock must be held. |
975 | **/ | 975 | **/ |
976 | int blk_queue_start_tag(request_queue_t *q, struct request *rq) | 976 | int blk_queue_start_tag(request_queue_t *q, struct request *rq) |
977 | { | 977 | { |
978 | struct blk_queue_tag *bqt = q->queue_tags; | 978 | struct blk_queue_tag *bqt = q->queue_tags; |
979 | int tag; | 979 | int tag; |
980 | 980 | ||
981 | if (unlikely((rq->flags & REQ_QUEUED))) { | 981 | if (unlikely((rq->flags & REQ_QUEUED))) { |
982 | printk(KERN_ERR | 982 | printk(KERN_ERR |
983 | "%s: request %p for device [%s] already tagged %d", | 983 | "%s: request %p for device [%s] already tagged %d", |
984 | __FUNCTION__, rq, | 984 | __FUNCTION__, rq, |
985 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); | 985 | rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); |
986 | BUG(); | 986 | BUG(); |
987 | } | 987 | } |
988 | 988 | ||
989 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); | 989 | tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); |
990 | if (tag >= bqt->max_depth) | 990 | if (tag >= bqt->max_depth) |
991 | return 1; | 991 | return 1; |
992 | 992 | ||
993 | __set_bit(tag, bqt->tag_map); | 993 | __set_bit(tag, bqt->tag_map); |
994 | 994 | ||
995 | rq->flags |= REQ_QUEUED; | 995 | rq->flags |= REQ_QUEUED; |
996 | rq->tag = tag; | 996 | rq->tag = tag; |
997 | bqt->tag_index[tag] = rq; | 997 | bqt->tag_index[tag] = rq; |
998 | blkdev_dequeue_request(rq); | 998 | blkdev_dequeue_request(rq); |
999 | list_add(&rq->queuelist, &bqt->busy_list); | 999 | list_add(&rq->queuelist, &bqt->busy_list); |
1000 | bqt->busy++; | 1000 | bqt->busy++; |
1001 | return 0; | 1001 | return 0; |
1002 | } | 1002 | } |
1003 | 1003 | ||
1004 | EXPORT_SYMBOL(blk_queue_start_tag); | 1004 | EXPORT_SYMBOL(blk_queue_start_tag); |
1005 | 1005 | ||
1006 | /** | 1006 | /** |
1007 | * blk_queue_invalidate_tags - invalidate all pending tags | 1007 | * blk_queue_invalidate_tags - invalidate all pending tags |
1008 | * @q: the request queue for the device | 1008 | * @q: the request queue for the device |
1009 | * | 1009 | * |
1010 | * Description: | 1010 | * Description: |
1011 | * Hardware conditions may dictate a need to stop all pending requests. | 1011 | * Hardware conditions may dictate a need to stop all pending requests. |
1012 | * In this case, we will safely clear the block side of the tag queue and | 1012 | * In this case, we will safely clear the block side of the tag queue and |
1013 | * readd all requests to the request queue in the right order. | 1013 | * readd all requests to the request queue in the right order. |
1014 | * | 1014 | * |
1015 | * Notes: | 1015 | * Notes: |
1016 | * queue lock must be held. | 1016 | * queue lock must be held. |
1017 | **/ | 1017 | **/ |
1018 | void blk_queue_invalidate_tags(request_queue_t *q) | 1018 | void blk_queue_invalidate_tags(request_queue_t *q) |
1019 | { | 1019 | { |
1020 | struct blk_queue_tag *bqt = q->queue_tags; | 1020 | struct blk_queue_tag *bqt = q->queue_tags; |
1021 | struct list_head *tmp, *n; | 1021 | struct list_head *tmp, *n; |
1022 | struct request *rq; | 1022 | struct request *rq; |
1023 | 1023 | ||
1024 | list_for_each_safe(tmp, n, &bqt->busy_list) { | 1024 | list_for_each_safe(tmp, n, &bqt->busy_list) { |
1025 | rq = list_entry_rq(tmp); | 1025 | rq = list_entry_rq(tmp); |
1026 | 1026 | ||
1027 | if (rq->tag == -1) { | 1027 | if (rq->tag == -1) { |
1028 | printk(KERN_ERR | 1028 | printk(KERN_ERR |
1029 | "%s: bad tag found on list\n", __FUNCTION__); | 1029 | "%s: bad tag found on list\n", __FUNCTION__); |
1030 | list_del_init(&rq->queuelist); | 1030 | list_del_init(&rq->queuelist); |
1031 | rq->flags &= ~REQ_QUEUED; | 1031 | rq->flags &= ~REQ_QUEUED; |
1032 | } else | 1032 | } else |
1033 | blk_queue_end_tag(q, rq); | 1033 | blk_queue_end_tag(q, rq); |
1034 | 1034 | ||
1035 | rq->flags &= ~REQ_STARTED; | 1035 | rq->flags &= ~REQ_STARTED; |
1036 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); | 1036 | __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); |
1037 | } | 1037 | } |
1038 | } | 1038 | } |
1039 | 1039 | ||
1040 | EXPORT_SYMBOL(blk_queue_invalidate_tags); | 1040 | EXPORT_SYMBOL(blk_queue_invalidate_tags); |
1041 | 1041 | ||
1042 | static char *rq_flags[] = { | 1042 | static char *rq_flags[] = { |
1043 | "REQ_RW", | 1043 | "REQ_RW", |
1044 | "REQ_FAILFAST", | 1044 | "REQ_FAILFAST", |
1045 | "REQ_SORTED", | 1045 | "REQ_SORTED", |
1046 | "REQ_SOFTBARRIER", | 1046 | "REQ_SOFTBARRIER", |
1047 | "REQ_HARDBARRIER", | 1047 | "REQ_HARDBARRIER", |
1048 | "REQ_CMD", | 1048 | "REQ_CMD", |
1049 | "REQ_NOMERGE", | 1049 | "REQ_NOMERGE", |
1050 | "REQ_STARTED", | 1050 | "REQ_STARTED", |
1051 | "REQ_DONTPREP", | 1051 | "REQ_DONTPREP", |
1052 | "REQ_QUEUED", | 1052 | "REQ_QUEUED", |
1053 | "REQ_ELVPRIV", | 1053 | "REQ_ELVPRIV", |
1054 | "REQ_PC", | 1054 | "REQ_PC", |
1055 | "REQ_BLOCK_PC", | 1055 | "REQ_BLOCK_PC", |
1056 | "REQ_SENSE", | 1056 | "REQ_SENSE", |
1057 | "REQ_FAILED", | 1057 | "REQ_FAILED", |
1058 | "REQ_QUIET", | 1058 | "REQ_QUIET", |
1059 | "REQ_SPECIAL", | 1059 | "REQ_SPECIAL", |
1060 | "REQ_DRIVE_CMD", | 1060 | "REQ_DRIVE_CMD", |
1061 | "REQ_DRIVE_TASK", | 1061 | "REQ_DRIVE_TASK", |
1062 | "REQ_DRIVE_TASKFILE", | 1062 | "REQ_DRIVE_TASKFILE", |
1063 | "REQ_PREEMPT", | 1063 | "REQ_PREEMPT", |
1064 | "REQ_PM_SUSPEND", | 1064 | "REQ_PM_SUSPEND", |
1065 | "REQ_PM_RESUME", | 1065 | "REQ_PM_RESUME", |
1066 | "REQ_PM_SHUTDOWN", | 1066 | "REQ_PM_SHUTDOWN", |
1067 | }; | 1067 | }; |
1068 | 1068 | ||
1069 | void blk_dump_rq_flags(struct request *rq, char *msg) | 1069 | void blk_dump_rq_flags(struct request *rq, char *msg) |
1070 | { | 1070 | { |
1071 | int bit; | 1071 | int bit; |
1072 | 1072 | ||
1073 | printk("%s: dev %s: flags = ", msg, | 1073 | printk("%s: dev %s: flags = ", msg, |
1074 | rq->rq_disk ? rq->rq_disk->disk_name : "?"); | 1074 | rq->rq_disk ? rq->rq_disk->disk_name : "?"); |
1075 | bit = 0; | 1075 | bit = 0; |
1076 | do { | 1076 | do { |
1077 | if (rq->flags & (1 << bit)) | 1077 | if (rq->flags & (1 << bit)) |
1078 | printk("%s ", rq_flags[bit]); | 1078 | printk("%s ", rq_flags[bit]); |
1079 | bit++; | 1079 | bit++; |
1080 | } while (bit < __REQ_NR_BITS); | 1080 | } while (bit < __REQ_NR_BITS); |
1081 | 1081 | ||
1082 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, | 1082 | printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, |
1083 | rq->nr_sectors, | 1083 | rq->nr_sectors, |
1084 | rq->current_nr_sectors); | 1084 | rq->current_nr_sectors); |
1085 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); | 1085 | printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); |
1086 | 1086 | ||
1087 | if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) { | 1087 | if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) { |
1088 | printk("cdb: "); | 1088 | printk("cdb: "); |
1089 | for (bit = 0; bit < sizeof(rq->cmd); bit++) | 1089 | for (bit = 0; bit < sizeof(rq->cmd); bit++) |
1090 | printk("%02x ", rq->cmd[bit]); | 1090 | printk("%02x ", rq->cmd[bit]); |
1091 | printk("\n"); | 1091 | printk("\n"); |
1092 | } | 1092 | } |
1093 | } | 1093 | } |
1094 | 1094 | ||
1095 | EXPORT_SYMBOL(blk_dump_rq_flags); | 1095 | EXPORT_SYMBOL(blk_dump_rq_flags); |
1096 | 1096 | ||
1097 | void blk_recount_segments(request_queue_t *q, struct bio *bio) | 1097 | void blk_recount_segments(request_queue_t *q, struct bio *bio) |
1098 | { | 1098 | { |
1099 | struct bio_vec *bv, *bvprv = NULL; | 1099 | struct bio_vec *bv, *bvprv = NULL; |
1100 | int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; | 1100 | int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; |
1101 | int high, highprv = 1; | 1101 | int high, highprv = 1; |
1102 | 1102 | ||
1103 | if (unlikely(!bio->bi_io_vec)) | 1103 | if (unlikely(!bio->bi_io_vec)) |
1104 | return; | 1104 | return; |
1105 | 1105 | ||
1106 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | 1106 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); |
1107 | hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; | 1107 | hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; |
1108 | bio_for_each_segment(bv, bio, i) { | 1108 | bio_for_each_segment(bv, bio, i) { |
1109 | /* | 1109 | /* |
1110 | * the trick here is making sure that a high page is never | 1110 | * the trick here is making sure that a high page is never |
1111 | * considered part of another segment, since that might | 1111 | * considered part of another segment, since that might |
1112 | * change with the bounce page. | 1112 | * change with the bounce page. |
1113 | */ | 1113 | */ |
1114 | high = page_to_pfn(bv->bv_page) >= q->bounce_pfn; | 1114 | high = page_to_pfn(bv->bv_page) >= q->bounce_pfn; |
1115 | if (high || highprv) | 1115 | if (high || highprv) |
1116 | goto new_hw_segment; | 1116 | goto new_hw_segment; |
1117 | if (cluster) { | 1117 | if (cluster) { |
1118 | if (seg_size + bv->bv_len > q->max_segment_size) | 1118 | if (seg_size + bv->bv_len > q->max_segment_size) |
1119 | goto new_segment; | 1119 | goto new_segment; |
1120 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) | 1120 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) |
1121 | goto new_segment; | 1121 | goto new_segment; |
1122 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) | 1122 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) |
1123 | goto new_segment; | 1123 | goto new_segment; |
1124 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) | 1124 | if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) |
1125 | goto new_hw_segment; | 1125 | goto new_hw_segment; |
1126 | 1126 | ||
1127 | seg_size += bv->bv_len; | 1127 | seg_size += bv->bv_len; |
1128 | hw_seg_size += bv->bv_len; | 1128 | hw_seg_size += bv->bv_len; |
1129 | bvprv = bv; | 1129 | bvprv = bv; |
1130 | continue; | 1130 | continue; |
1131 | } | 1131 | } |
1132 | new_segment: | 1132 | new_segment: |
1133 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && | 1133 | if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && |
1134 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { | 1134 | !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { |
1135 | hw_seg_size += bv->bv_len; | 1135 | hw_seg_size += bv->bv_len; |
1136 | } else { | 1136 | } else { |
1137 | new_hw_segment: | 1137 | new_hw_segment: |
1138 | if (hw_seg_size > bio->bi_hw_front_size) | 1138 | if (hw_seg_size > bio->bi_hw_front_size) |
1139 | bio->bi_hw_front_size = hw_seg_size; | 1139 | bio->bi_hw_front_size = hw_seg_size; |
1140 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; | 1140 | hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; |
1141 | nr_hw_segs++; | 1141 | nr_hw_segs++; |
1142 | } | 1142 | } |
1143 | 1143 | ||
1144 | nr_phys_segs++; | 1144 | nr_phys_segs++; |
1145 | bvprv = bv; | 1145 | bvprv = bv; |
1146 | seg_size = bv->bv_len; | 1146 | seg_size = bv->bv_len; |
1147 | highprv = high; | 1147 | highprv = high; |
1148 | } | 1148 | } |
1149 | if (hw_seg_size > bio->bi_hw_back_size) | 1149 | if (hw_seg_size > bio->bi_hw_back_size) |
1150 | bio->bi_hw_back_size = hw_seg_size; | 1150 | bio->bi_hw_back_size = hw_seg_size; |
1151 | if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) | 1151 | if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) |
1152 | bio->bi_hw_front_size = hw_seg_size; | 1152 | bio->bi_hw_front_size = hw_seg_size; |
1153 | bio->bi_phys_segments = nr_phys_segs; | 1153 | bio->bi_phys_segments = nr_phys_segs; |
1154 | bio->bi_hw_segments = nr_hw_segs; | 1154 | bio->bi_hw_segments = nr_hw_segs; |
1155 | bio->bi_flags |= (1 << BIO_SEG_VALID); | 1155 | bio->bi_flags |= (1 << BIO_SEG_VALID); |
1156 | } | 1156 | } |
1157 | 1157 | ||
1158 | 1158 | ||
1159 | static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, | 1159 | static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, |
1160 | struct bio *nxt) | 1160 | struct bio *nxt) |
1161 | { | 1161 | { |
1162 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) | 1162 | if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) |
1163 | return 0; | 1163 | return 0; |
1164 | 1164 | ||
1165 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) | 1165 | if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) |
1166 | return 0; | 1166 | return 0; |
1167 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | 1167 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) |
1168 | return 0; | 1168 | return 0; |
1169 | 1169 | ||
1170 | /* | 1170 | /* |
1171 | * bio and nxt are contigous in memory, check if the queue allows | 1171 | * bio and nxt are contigous in memory, check if the queue allows |
1172 | * these two to be merged into one | 1172 | * these two to be merged into one |
1173 | */ | 1173 | */ |
1174 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) | 1174 | if (BIO_SEG_BOUNDARY(q, bio, nxt)) |
1175 | return 1; | 1175 | return 1; |
1176 | 1176 | ||
1177 | return 0; | 1177 | return 0; |
1178 | } | 1178 | } |
1179 | 1179 | ||
1180 | static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio, | 1180 | static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio, |
1181 | struct bio *nxt) | 1181 | struct bio *nxt) |
1182 | { | 1182 | { |
1183 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | 1183 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
1184 | blk_recount_segments(q, bio); | 1184 | blk_recount_segments(q, bio); |
1185 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) | 1185 | if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) |
1186 | blk_recount_segments(q, nxt); | 1186 | blk_recount_segments(q, nxt); |
1187 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || | 1187 | if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || |
1188 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size)) | 1188 | BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size)) |
1189 | return 0; | 1189 | return 0; |
1190 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) | 1190 | if (bio->bi_size + nxt->bi_size > q->max_segment_size) |
1191 | return 0; | 1191 | return 0; |
1192 | 1192 | ||
1193 | return 1; | 1193 | return 1; |
1194 | } | 1194 | } |
1195 | 1195 | ||
1196 | /* | 1196 | /* |
1197 | * map a request to scatterlist, return number of sg entries setup. Caller | 1197 | * map a request to scatterlist, return number of sg entries setup. Caller |
1198 | * must make sure sg can hold rq->nr_phys_segments entries | 1198 | * must make sure sg can hold rq->nr_phys_segments entries |
1199 | */ | 1199 | */ |
1200 | int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg) | 1200 | int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg) |
1201 | { | 1201 | { |
1202 | struct bio_vec *bvec, *bvprv; | 1202 | struct bio_vec *bvec, *bvprv; |
1203 | struct bio *bio; | 1203 | struct bio *bio; |
1204 | int nsegs, i, cluster; | 1204 | int nsegs, i, cluster; |
1205 | 1205 | ||
1206 | nsegs = 0; | 1206 | nsegs = 0; |
1207 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); | 1207 | cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); |
1208 | 1208 | ||
1209 | /* | 1209 | /* |
1210 | * for each bio in rq | 1210 | * for each bio in rq |
1211 | */ | 1211 | */ |
1212 | bvprv = NULL; | 1212 | bvprv = NULL; |
1213 | rq_for_each_bio(bio, rq) { | 1213 | rq_for_each_bio(bio, rq) { |
1214 | /* | 1214 | /* |
1215 | * for each segment in bio | 1215 | * for each segment in bio |
1216 | */ | 1216 | */ |
1217 | bio_for_each_segment(bvec, bio, i) { | 1217 | bio_for_each_segment(bvec, bio, i) { |
1218 | int nbytes = bvec->bv_len; | 1218 | int nbytes = bvec->bv_len; |
1219 | 1219 | ||
1220 | if (bvprv && cluster) { | 1220 | if (bvprv && cluster) { |
1221 | if (sg[nsegs - 1].length + nbytes > q->max_segment_size) | 1221 | if (sg[nsegs - 1].length + nbytes > q->max_segment_size) |
1222 | goto new_segment; | 1222 | goto new_segment; |
1223 | 1223 | ||
1224 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) | 1224 | if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) |
1225 | goto new_segment; | 1225 | goto new_segment; |
1226 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) | 1226 | if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) |
1227 | goto new_segment; | 1227 | goto new_segment; |
1228 | 1228 | ||
1229 | sg[nsegs - 1].length += nbytes; | 1229 | sg[nsegs - 1].length += nbytes; |
1230 | } else { | 1230 | } else { |
1231 | new_segment: | 1231 | new_segment: |
1232 | memset(&sg[nsegs],0,sizeof(struct scatterlist)); | 1232 | memset(&sg[nsegs],0,sizeof(struct scatterlist)); |
1233 | sg[nsegs].page = bvec->bv_page; | 1233 | sg[nsegs].page = bvec->bv_page; |
1234 | sg[nsegs].length = nbytes; | 1234 | sg[nsegs].length = nbytes; |
1235 | sg[nsegs].offset = bvec->bv_offset; | 1235 | sg[nsegs].offset = bvec->bv_offset; |
1236 | 1236 | ||
1237 | nsegs++; | 1237 | nsegs++; |
1238 | } | 1238 | } |
1239 | bvprv = bvec; | 1239 | bvprv = bvec; |
1240 | } /* segments in bio */ | 1240 | } /* segments in bio */ |
1241 | } /* bios in rq */ | 1241 | } /* bios in rq */ |
1242 | 1242 | ||
1243 | return nsegs; | 1243 | return nsegs; |
1244 | } | 1244 | } |
1245 | 1245 | ||
1246 | EXPORT_SYMBOL(blk_rq_map_sg); | 1246 | EXPORT_SYMBOL(blk_rq_map_sg); |
1247 | 1247 | ||
1248 | /* | 1248 | /* |
1249 | * the standard queue merge functions, can be overridden with device | 1249 | * the standard queue merge functions, can be overridden with device |
1250 | * specific ones if so desired | 1250 | * specific ones if so desired |
1251 | */ | 1251 | */ |
1252 | 1252 | ||
1253 | static inline int ll_new_mergeable(request_queue_t *q, | 1253 | static inline int ll_new_mergeable(request_queue_t *q, |
1254 | struct request *req, | 1254 | struct request *req, |
1255 | struct bio *bio) | 1255 | struct bio *bio) |
1256 | { | 1256 | { |
1257 | int nr_phys_segs = bio_phys_segments(q, bio); | 1257 | int nr_phys_segs = bio_phys_segments(q, bio); |
1258 | 1258 | ||
1259 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | 1259 | if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { |
1260 | req->flags |= REQ_NOMERGE; | 1260 | req->flags |= REQ_NOMERGE; |
1261 | if (req == q->last_merge) | 1261 | if (req == q->last_merge) |
1262 | q->last_merge = NULL; | 1262 | q->last_merge = NULL; |
1263 | return 0; | 1263 | return 0; |
1264 | } | 1264 | } |
1265 | 1265 | ||
1266 | /* | 1266 | /* |
1267 | * A hw segment is just getting larger, bump just the phys | 1267 | * A hw segment is just getting larger, bump just the phys |
1268 | * counter. | 1268 | * counter. |
1269 | */ | 1269 | */ |
1270 | req->nr_phys_segments += nr_phys_segs; | 1270 | req->nr_phys_segments += nr_phys_segs; |
1271 | return 1; | 1271 | return 1; |
1272 | } | 1272 | } |
1273 | 1273 | ||
1274 | static inline int ll_new_hw_segment(request_queue_t *q, | 1274 | static inline int ll_new_hw_segment(request_queue_t *q, |
1275 | struct request *req, | 1275 | struct request *req, |
1276 | struct bio *bio) | 1276 | struct bio *bio) |
1277 | { | 1277 | { |
1278 | int nr_hw_segs = bio_hw_segments(q, bio); | 1278 | int nr_hw_segs = bio_hw_segments(q, bio); |
1279 | int nr_phys_segs = bio_phys_segments(q, bio); | 1279 | int nr_phys_segs = bio_phys_segments(q, bio); |
1280 | 1280 | ||
1281 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments | 1281 | if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments |
1282 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { | 1282 | || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { |
1283 | req->flags |= REQ_NOMERGE; | 1283 | req->flags |= REQ_NOMERGE; |
1284 | if (req == q->last_merge) | 1284 | if (req == q->last_merge) |
1285 | q->last_merge = NULL; | 1285 | q->last_merge = NULL; |
1286 | return 0; | 1286 | return 0; |
1287 | } | 1287 | } |
1288 | 1288 | ||
1289 | /* | 1289 | /* |
1290 | * This will form the start of a new hw segment. Bump both | 1290 | * This will form the start of a new hw segment. Bump both |
1291 | * counters. | 1291 | * counters. |
1292 | */ | 1292 | */ |
1293 | req->nr_hw_segments += nr_hw_segs; | 1293 | req->nr_hw_segments += nr_hw_segs; |
1294 | req->nr_phys_segments += nr_phys_segs; | 1294 | req->nr_phys_segments += nr_phys_segs; |
1295 | return 1; | 1295 | return 1; |
1296 | } | 1296 | } |
1297 | 1297 | ||
1298 | static int ll_back_merge_fn(request_queue_t *q, struct request *req, | 1298 | static int ll_back_merge_fn(request_queue_t *q, struct request *req, |
1299 | struct bio *bio) | 1299 | struct bio *bio) |
1300 | { | 1300 | { |
1301 | unsigned short max_sectors; | 1301 | unsigned short max_sectors; |
1302 | int len; | 1302 | int len; |
1303 | 1303 | ||
1304 | if (unlikely(blk_pc_request(req))) | 1304 | if (unlikely(blk_pc_request(req))) |
1305 | max_sectors = q->max_hw_sectors; | 1305 | max_sectors = q->max_hw_sectors; |
1306 | else | 1306 | else |
1307 | max_sectors = q->max_sectors; | 1307 | max_sectors = q->max_sectors; |
1308 | 1308 | ||
1309 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | 1309 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { |
1310 | req->flags |= REQ_NOMERGE; | 1310 | req->flags |= REQ_NOMERGE; |
1311 | if (req == q->last_merge) | 1311 | if (req == q->last_merge) |
1312 | q->last_merge = NULL; | 1312 | q->last_merge = NULL; |
1313 | return 0; | 1313 | return 0; |
1314 | } | 1314 | } |
1315 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) | 1315 | if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) |
1316 | blk_recount_segments(q, req->biotail); | 1316 | blk_recount_segments(q, req->biotail); |
1317 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | 1317 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
1318 | blk_recount_segments(q, bio); | 1318 | blk_recount_segments(q, bio); |
1319 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; | 1319 | len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; |
1320 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && | 1320 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && |
1321 | !BIOVEC_VIRT_OVERSIZE(len)) { | 1321 | !BIOVEC_VIRT_OVERSIZE(len)) { |
1322 | int mergeable = ll_new_mergeable(q, req, bio); | 1322 | int mergeable = ll_new_mergeable(q, req, bio); |
1323 | 1323 | ||
1324 | if (mergeable) { | 1324 | if (mergeable) { |
1325 | if (req->nr_hw_segments == 1) | 1325 | if (req->nr_hw_segments == 1) |
1326 | req->bio->bi_hw_front_size = len; | 1326 | req->bio->bi_hw_front_size = len; |
1327 | if (bio->bi_hw_segments == 1) | 1327 | if (bio->bi_hw_segments == 1) |
1328 | bio->bi_hw_back_size = len; | 1328 | bio->bi_hw_back_size = len; |
1329 | } | 1329 | } |
1330 | return mergeable; | 1330 | return mergeable; |
1331 | } | 1331 | } |
1332 | 1332 | ||
1333 | return ll_new_hw_segment(q, req, bio); | 1333 | return ll_new_hw_segment(q, req, bio); |
1334 | } | 1334 | } |
1335 | 1335 | ||
1336 | static int ll_front_merge_fn(request_queue_t *q, struct request *req, | 1336 | static int ll_front_merge_fn(request_queue_t *q, struct request *req, |
1337 | struct bio *bio) | 1337 | struct bio *bio) |
1338 | { | 1338 | { |
1339 | unsigned short max_sectors; | 1339 | unsigned short max_sectors; |
1340 | int len; | 1340 | int len; |
1341 | 1341 | ||
1342 | if (unlikely(blk_pc_request(req))) | 1342 | if (unlikely(blk_pc_request(req))) |
1343 | max_sectors = q->max_hw_sectors; | 1343 | max_sectors = q->max_hw_sectors; |
1344 | else | 1344 | else |
1345 | max_sectors = q->max_sectors; | 1345 | max_sectors = q->max_sectors; |
1346 | 1346 | ||
1347 | 1347 | ||
1348 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { | 1348 | if (req->nr_sectors + bio_sectors(bio) > max_sectors) { |
1349 | req->flags |= REQ_NOMERGE; | 1349 | req->flags |= REQ_NOMERGE; |
1350 | if (req == q->last_merge) | 1350 | if (req == q->last_merge) |
1351 | q->last_merge = NULL; | 1351 | q->last_merge = NULL; |
1352 | return 0; | 1352 | return 0; |
1353 | } | 1353 | } |
1354 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; | 1354 | len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; |
1355 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) | 1355 | if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) |
1356 | blk_recount_segments(q, bio); | 1356 | blk_recount_segments(q, bio); |
1357 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) | 1357 | if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) |
1358 | blk_recount_segments(q, req->bio); | 1358 | blk_recount_segments(q, req->bio); |
1359 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && | 1359 | if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && |
1360 | !BIOVEC_VIRT_OVERSIZE(len)) { | 1360 | !BIOVEC_VIRT_OVERSIZE(len)) { |
1361 | int mergeable = ll_new_mergeable(q, req, bio); | 1361 | int mergeable = ll_new_mergeable(q, req, bio); |
1362 | 1362 | ||
1363 | if (mergeable) { | 1363 | if (mergeable) { |
1364 | if (bio->bi_hw_segments == 1) | 1364 | if (bio->bi_hw_segments == 1) |
1365 | bio->bi_hw_front_size = len; | 1365 | bio->bi_hw_front_size = len; |
1366 | if (req->nr_hw_segments == 1) | 1366 | if (req->nr_hw_segments == 1) |
1367 | req->biotail->bi_hw_back_size = len; | 1367 | req->biotail->bi_hw_back_size = len; |
1368 | } | 1368 | } |
1369 | return mergeable; | 1369 | return mergeable; |
1370 | } | 1370 | } |
1371 | 1371 | ||
1372 | return ll_new_hw_segment(q, req, bio); | 1372 | return ll_new_hw_segment(q, req, bio); |
1373 | } | 1373 | } |
1374 | 1374 | ||
1375 | static int ll_merge_requests_fn(request_queue_t *q, struct request *req, | 1375 | static int ll_merge_requests_fn(request_queue_t *q, struct request *req, |
1376 | struct request *next) | 1376 | struct request *next) |
1377 | { | 1377 | { |
1378 | int total_phys_segments; | 1378 | int total_phys_segments; |
1379 | int total_hw_segments; | 1379 | int total_hw_segments; |
1380 | 1380 | ||
1381 | /* | 1381 | /* |
1382 | * First check if the either of the requests are re-queued | 1382 | * First check if the either of the requests are re-queued |
1383 | * requests. Can't merge them if they are. | 1383 | * requests. Can't merge them if they are. |
1384 | */ | 1384 | */ |
1385 | if (req->special || next->special) | 1385 | if (req->special || next->special) |
1386 | return 0; | 1386 | return 0; |
1387 | 1387 | ||
1388 | /* | 1388 | /* |
1389 | * Will it become too large? | 1389 | * Will it become too large? |
1390 | */ | 1390 | */ |
1391 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) | 1391 | if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) |
1392 | return 0; | 1392 | return 0; |
1393 | 1393 | ||
1394 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; | 1394 | total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; |
1395 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) | 1395 | if (blk_phys_contig_segment(q, req->biotail, next->bio)) |
1396 | total_phys_segments--; | 1396 | total_phys_segments--; |
1397 | 1397 | ||
1398 | if (total_phys_segments > q->max_phys_segments) | 1398 | if (total_phys_segments > q->max_phys_segments) |
1399 | return 0; | 1399 | return 0; |
1400 | 1400 | ||
1401 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; | 1401 | total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; |
1402 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { | 1402 | if (blk_hw_contig_segment(q, req->biotail, next->bio)) { |
1403 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; | 1403 | int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; |
1404 | /* | 1404 | /* |
1405 | * propagate the combined length to the end of the requests | 1405 | * propagate the combined length to the end of the requests |
1406 | */ | 1406 | */ |
1407 | if (req->nr_hw_segments == 1) | 1407 | if (req->nr_hw_segments == 1) |
1408 | req->bio->bi_hw_front_size = len; | 1408 | req->bio->bi_hw_front_size = len; |
1409 | if (next->nr_hw_segments == 1) | 1409 | if (next->nr_hw_segments == 1) |
1410 | next->biotail->bi_hw_back_size = len; | 1410 | next->biotail->bi_hw_back_size = len; |
1411 | total_hw_segments--; | 1411 | total_hw_segments--; |
1412 | } | 1412 | } |
1413 | 1413 | ||
1414 | if (total_hw_segments > q->max_hw_segments) | 1414 | if (total_hw_segments > q->max_hw_segments) |
1415 | return 0; | 1415 | return 0; |
1416 | 1416 | ||
1417 | /* Merge is OK... */ | 1417 | /* Merge is OK... */ |
1418 | req->nr_phys_segments = total_phys_segments; | 1418 | req->nr_phys_segments = total_phys_segments; |
1419 | req->nr_hw_segments = total_hw_segments; | 1419 | req->nr_hw_segments = total_hw_segments; |
1420 | return 1; | 1420 | return 1; |
1421 | } | 1421 | } |
1422 | 1422 | ||
1423 | /* | 1423 | /* |
1424 | * "plug" the device if there are no outstanding requests: this will | 1424 | * "plug" the device if there are no outstanding requests: this will |
1425 | * force the transfer to start only after we have put all the requests | 1425 | * force the transfer to start only after we have put all the requests |
1426 | * on the list. | 1426 | * on the list. |
1427 | * | 1427 | * |
1428 | * This is called with interrupts off and no requests on the queue and | 1428 | * This is called with interrupts off and no requests on the queue and |
1429 | * with the queue lock held. | 1429 | * with the queue lock held. |
1430 | */ | 1430 | */ |
1431 | void blk_plug_device(request_queue_t *q) | 1431 | void blk_plug_device(request_queue_t *q) |
1432 | { | 1432 | { |
1433 | WARN_ON(!irqs_disabled()); | 1433 | WARN_ON(!irqs_disabled()); |
1434 | 1434 | ||
1435 | /* | 1435 | /* |
1436 | * don't plug a stopped queue, it must be paired with blk_start_queue() | 1436 | * don't plug a stopped queue, it must be paired with blk_start_queue() |
1437 | * which will restart the queueing | 1437 | * which will restart the queueing |
1438 | */ | 1438 | */ |
1439 | if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)) | 1439 | if (test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags)) |
1440 | return; | 1440 | return; |
1441 | 1441 | ||
1442 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | 1442 | if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) |
1443 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); | 1443 | mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); |
1444 | } | 1444 | } |
1445 | 1445 | ||
1446 | EXPORT_SYMBOL(blk_plug_device); | 1446 | EXPORT_SYMBOL(blk_plug_device); |
1447 | 1447 | ||
1448 | /* | 1448 | /* |
1449 | * remove the queue from the plugged list, if present. called with | 1449 | * remove the queue from the plugged list, if present. called with |
1450 | * queue lock held and interrupts disabled. | 1450 | * queue lock held and interrupts disabled. |
1451 | */ | 1451 | */ |
1452 | int blk_remove_plug(request_queue_t *q) | 1452 | int blk_remove_plug(request_queue_t *q) |
1453 | { | 1453 | { |
1454 | WARN_ON(!irqs_disabled()); | 1454 | WARN_ON(!irqs_disabled()); |
1455 | 1455 | ||
1456 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) | 1456 | if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) |
1457 | return 0; | 1457 | return 0; |
1458 | 1458 | ||
1459 | del_timer(&q->unplug_timer); | 1459 | del_timer(&q->unplug_timer); |
1460 | return 1; | 1460 | return 1; |
1461 | } | 1461 | } |
1462 | 1462 | ||
1463 | EXPORT_SYMBOL(blk_remove_plug); | 1463 | EXPORT_SYMBOL(blk_remove_plug); |
1464 | 1464 | ||
1465 | /* | 1465 | /* |
1466 | * remove the plug and let it rip.. | 1466 | * remove the plug and let it rip.. |
1467 | */ | 1467 | */ |
1468 | void __generic_unplug_device(request_queue_t *q) | 1468 | void __generic_unplug_device(request_queue_t *q) |
1469 | { | 1469 | { |
1470 | if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))) | 1470 | if (unlikely(test_bit(QUEUE_FLAG_STOPPED, &q->queue_flags))) |
1471 | return; | 1471 | return; |
1472 | 1472 | ||
1473 | if (!blk_remove_plug(q)) | 1473 | if (!blk_remove_plug(q)) |
1474 | return; | 1474 | return; |
1475 | 1475 | ||
1476 | q->request_fn(q); | 1476 | q->request_fn(q); |
1477 | } | 1477 | } |
1478 | EXPORT_SYMBOL(__generic_unplug_device); | 1478 | EXPORT_SYMBOL(__generic_unplug_device); |
1479 | 1479 | ||
1480 | /** | 1480 | /** |
1481 | * generic_unplug_device - fire a request queue | 1481 | * generic_unplug_device - fire a request queue |
1482 | * @q: The &request_queue_t in question | 1482 | * @q: The &request_queue_t in question |
1483 | * | 1483 | * |
1484 | * Description: | 1484 | * Description: |
1485 | * Linux uses plugging to build bigger requests queues before letting | 1485 | * Linux uses plugging to build bigger requests queues before letting |
1486 | * the device have at them. If a queue is plugged, the I/O scheduler | 1486 | * the device have at them. If a queue is plugged, the I/O scheduler |
1487 | * is still adding and merging requests on the queue. Once the queue | 1487 | * is still adding and merging requests on the queue. Once the queue |
1488 | * gets unplugged, the request_fn defined for the queue is invoked and | 1488 | * gets unplugged, the request_fn defined for the queue is invoked and |
1489 | * transfers started. | 1489 | * transfers started. |
1490 | **/ | 1490 | **/ |
1491 | void generic_unplug_device(request_queue_t *q) | 1491 | void generic_unplug_device(request_queue_t *q) |
1492 | { | 1492 | { |
1493 | spin_lock_irq(q->queue_lock); | 1493 | spin_lock_irq(q->queue_lock); |
1494 | __generic_unplug_device(q); | 1494 | __generic_unplug_device(q); |
1495 | spin_unlock_irq(q->queue_lock); | 1495 | spin_unlock_irq(q->queue_lock); |
1496 | } | 1496 | } |
1497 | EXPORT_SYMBOL(generic_unplug_device); | 1497 | EXPORT_SYMBOL(generic_unplug_device); |
1498 | 1498 | ||
1499 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, | 1499 | static void blk_backing_dev_unplug(struct backing_dev_info *bdi, |
1500 | struct page *page) | 1500 | struct page *page) |
1501 | { | 1501 | { |
1502 | request_queue_t *q = bdi->unplug_io_data; | 1502 | request_queue_t *q = bdi->unplug_io_data; |
1503 | 1503 | ||
1504 | /* | 1504 | /* |
1505 | * devices don't necessarily have an ->unplug_fn defined | 1505 | * devices don't necessarily have an ->unplug_fn defined |
1506 | */ | 1506 | */ |
1507 | if (q->unplug_fn) | 1507 | if (q->unplug_fn) |
1508 | q->unplug_fn(q); | 1508 | q->unplug_fn(q); |
1509 | } | 1509 | } |
1510 | 1510 | ||
1511 | static void blk_unplug_work(void *data) | 1511 | static void blk_unplug_work(void *data) |
1512 | { | 1512 | { |
1513 | request_queue_t *q = data; | 1513 | request_queue_t *q = data; |
1514 | 1514 | ||
1515 | q->unplug_fn(q); | 1515 | q->unplug_fn(q); |
1516 | } | 1516 | } |
1517 | 1517 | ||
1518 | static void blk_unplug_timeout(unsigned long data) | 1518 | static void blk_unplug_timeout(unsigned long data) |
1519 | { | 1519 | { |
1520 | request_queue_t *q = (request_queue_t *)data; | 1520 | request_queue_t *q = (request_queue_t *)data; |
1521 | 1521 | ||
1522 | kblockd_schedule_work(&q->unplug_work); | 1522 | kblockd_schedule_work(&q->unplug_work); |
1523 | } | 1523 | } |
1524 | 1524 | ||
1525 | /** | 1525 | /** |
1526 | * blk_start_queue - restart a previously stopped queue | 1526 | * blk_start_queue - restart a previously stopped queue |
1527 | * @q: The &request_queue_t in question | 1527 | * @q: The &request_queue_t in question |
1528 | * | 1528 | * |
1529 | * Description: | 1529 | * Description: |
1530 | * blk_start_queue() will clear the stop flag on the queue, and call | 1530 | * blk_start_queue() will clear the stop flag on the queue, and call |
1531 | * the request_fn for the queue if it was in a stopped state when | 1531 | * the request_fn for the queue if it was in a stopped state when |
1532 | * entered. Also see blk_stop_queue(). Queue lock must be held. | 1532 | * entered. Also see blk_stop_queue(). Queue lock must be held. |
1533 | **/ | 1533 | **/ |
1534 | void blk_start_queue(request_queue_t *q) | 1534 | void blk_start_queue(request_queue_t *q) |
1535 | { | 1535 | { |
1536 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | 1536 | clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
1537 | 1537 | ||
1538 | /* | 1538 | /* |
1539 | * one level of recursion is ok and is much faster than kicking | 1539 | * one level of recursion is ok and is much faster than kicking |
1540 | * the unplug handling | 1540 | * the unplug handling |
1541 | */ | 1541 | */ |
1542 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { | 1542 | if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { |
1543 | q->request_fn(q); | 1543 | q->request_fn(q); |
1544 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); | 1544 | clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); |
1545 | } else { | 1545 | } else { |
1546 | blk_plug_device(q); | 1546 | blk_plug_device(q); |
1547 | kblockd_schedule_work(&q->unplug_work); | 1547 | kblockd_schedule_work(&q->unplug_work); |
1548 | } | 1548 | } |
1549 | } | 1549 | } |
1550 | 1550 | ||
1551 | EXPORT_SYMBOL(blk_start_queue); | 1551 | EXPORT_SYMBOL(blk_start_queue); |
1552 | 1552 | ||
1553 | /** | 1553 | /** |
1554 | * blk_stop_queue - stop a queue | 1554 | * blk_stop_queue - stop a queue |
1555 | * @q: The &request_queue_t in question | 1555 | * @q: The &request_queue_t in question |
1556 | * | 1556 | * |
1557 | * Description: | 1557 | * Description: |
1558 | * The Linux block layer assumes that a block driver will consume all | 1558 | * The Linux block layer assumes that a block driver will consume all |
1559 | * entries on the request queue when the request_fn strategy is called. | 1559 | * entries on the request queue when the request_fn strategy is called. |
1560 | * Often this will not happen, because of hardware limitations (queue | 1560 | * Often this will not happen, because of hardware limitations (queue |
1561 | * depth settings). If a device driver gets a 'queue full' response, | 1561 | * depth settings). If a device driver gets a 'queue full' response, |
1562 | * or if it simply chooses not to queue more I/O at one point, it can | 1562 | * or if it simply chooses not to queue more I/O at one point, it can |
1563 | * call this function to prevent the request_fn from being called until | 1563 | * call this function to prevent the request_fn from being called until |
1564 | * the driver has signalled it's ready to go again. This happens by calling | 1564 | * the driver has signalled it's ready to go again. This happens by calling |
1565 | * blk_start_queue() to restart queue operations. Queue lock must be held. | 1565 | * blk_start_queue() to restart queue operations. Queue lock must be held. |
1566 | **/ | 1566 | **/ |
1567 | void blk_stop_queue(request_queue_t *q) | 1567 | void blk_stop_queue(request_queue_t *q) |
1568 | { | 1568 | { |
1569 | blk_remove_plug(q); | 1569 | blk_remove_plug(q); |
1570 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); | 1570 | set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); |
1571 | } | 1571 | } |
1572 | EXPORT_SYMBOL(blk_stop_queue); | 1572 | EXPORT_SYMBOL(blk_stop_queue); |
1573 | 1573 | ||
1574 | /** | 1574 | /** |
1575 | * blk_sync_queue - cancel any pending callbacks on a queue | 1575 | * blk_sync_queue - cancel any pending callbacks on a queue |
1576 | * @q: the queue | 1576 | * @q: the queue |
1577 | * | 1577 | * |
1578 | * Description: | 1578 | * Description: |
1579 | * The block layer may perform asynchronous callback activity | 1579 | * The block layer may perform asynchronous callback activity |
1580 | * on a queue, such as calling the unplug function after a timeout. | 1580 | * on a queue, such as calling the unplug function after a timeout. |
1581 | * A block device may call blk_sync_queue to ensure that any | 1581 | * A block device may call blk_sync_queue to ensure that any |
1582 | * such activity is cancelled, thus allowing it to release resources | 1582 | * such activity is cancelled, thus allowing it to release resources |
1583 | * the the callbacks might use. The caller must already have made sure | 1583 | * the the callbacks might use. The caller must already have made sure |
1584 | * that its ->make_request_fn will not re-add plugging prior to calling | 1584 | * that its ->make_request_fn will not re-add plugging prior to calling |
1585 | * this function. | 1585 | * this function. |
1586 | * | 1586 | * |
1587 | */ | 1587 | */ |
1588 | void blk_sync_queue(struct request_queue *q) | 1588 | void blk_sync_queue(struct request_queue *q) |
1589 | { | 1589 | { |
1590 | del_timer_sync(&q->unplug_timer); | 1590 | del_timer_sync(&q->unplug_timer); |
1591 | kblockd_flush(); | 1591 | kblockd_flush(); |
1592 | } | 1592 | } |
1593 | EXPORT_SYMBOL(blk_sync_queue); | 1593 | EXPORT_SYMBOL(blk_sync_queue); |
1594 | 1594 | ||
1595 | /** | 1595 | /** |
1596 | * blk_run_queue - run a single device queue | 1596 | * blk_run_queue - run a single device queue |
1597 | * @q: The queue to run | 1597 | * @q: The queue to run |
1598 | */ | 1598 | */ |
1599 | void blk_run_queue(struct request_queue *q) | 1599 | void blk_run_queue(struct request_queue *q) |
1600 | { | 1600 | { |
1601 | unsigned long flags; | 1601 | unsigned long flags; |
1602 | 1602 | ||
1603 | spin_lock_irqsave(q->queue_lock, flags); | 1603 | spin_lock_irqsave(q->queue_lock, flags); |
1604 | blk_remove_plug(q); | 1604 | blk_remove_plug(q); |
1605 | if (!elv_queue_empty(q)) | 1605 | if (!elv_queue_empty(q)) |
1606 | q->request_fn(q); | 1606 | q->request_fn(q); |
1607 | spin_unlock_irqrestore(q->queue_lock, flags); | 1607 | spin_unlock_irqrestore(q->queue_lock, flags); |
1608 | } | 1608 | } |
1609 | EXPORT_SYMBOL(blk_run_queue); | 1609 | EXPORT_SYMBOL(blk_run_queue); |
1610 | 1610 | ||
1611 | /** | 1611 | /** |
1612 | * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed | 1612 | * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed |
1613 | * @q: the request queue to be released | 1613 | * @q: the request queue to be released |
1614 | * | 1614 | * |
1615 | * Description: | 1615 | * Description: |
1616 | * blk_cleanup_queue is the pair to blk_init_queue() or | 1616 | * blk_cleanup_queue is the pair to blk_init_queue() or |
1617 | * blk_queue_make_request(). It should be called when a request queue is | 1617 | * blk_queue_make_request(). It should be called when a request queue is |
1618 | * being released; typically when a block device is being de-registered. | 1618 | * being released; typically when a block device is being de-registered. |
1619 | * Currently, its primary task it to free all the &struct request | 1619 | * Currently, its primary task it to free all the &struct request |
1620 | * structures that were allocated to the queue and the queue itself. | 1620 | * structures that were allocated to the queue and the queue itself. |
1621 | * | 1621 | * |
1622 | * Caveat: | 1622 | * Caveat: |
1623 | * Hopefully the low level driver will have finished any | 1623 | * Hopefully the low level driver will have finished any |
1624 | * outstanding requests first... | 1624 | * outstanding requests first... |
1625 | **/ | 1625 | **/ |
1626 | void blk_cleanup_queue(request_queue_t * q) | 1626 | void blk_cleanup_queue(request_queue_t * q) |
1627 | { | 1627 | { |
1628 | struct request_list *rl = &q->rq; | 1628 | struct request_list *rl = &q->rq; |
1629 | 1629 | ||
1630 | if (!atomic_dec_and_test(&q->refcnt)) | 1630 | if (!atomic_dec_and_test(&q->refcnt)) |
1631 | return; | 1631 | return; |
1632 | 1632 | ||
1633 | if (q->elevator) | 1633 | if (q->elevator) |
1634 | elevator_exit(q->elevator); | 1634 | elevator_exit(q->elevator); |
1635 | 1635 | ||
1636 | blk_sync_queue(q); | 1636 | blk_sync_queue(q); |
1637 | 1637 | ||
1638 | if (rl->rq_pool) | 1638 | if (rl->rq_pool) |
1639 | mempool_destroy(rl->rq_pool); | 1639 | mempool_destroy(rl->rq_pool); |
1640 | 1640 | ||
1641 | if (q->queue_tags) | 1641 | if (q->queue_tags) |
1642 | __blk_queue_free_tags(q); | 1642 | __blk_queue_free_tags(q); |
1643 | 1643 | ||
1644 | blk_queue_ordered(q, QUEUE_ORDERED_NONE); | 1644 | blk_queue_ordered(q, QUEUE_ORDERED_NONE); |
1645 | 1645 | ||
1646 | kmem_cache_free(requestq_cachep, q); | 1646 | kmem_cache_free(requestq_cachep, q); |
1647 | } | 1647 | } |
1648 | 1648 | ||
1649 | EXPORT_SYMBOL(blk_cleanup_queue); | 1649 | EXPORT_SYMBOL(blk_cleanup_queue); |
1650 | 1650 | ||
1651 | static int blk_init_free_list(request_queue_t *q) | 1651 | static int blk_init_free_list(request_queue_t *q) |
1652 | { | 1652 | { |
1653 | struct request_list *rl = &q->rq; | 1653 | struct request_list *rl = &q->rq; |
1654 | 1654 | ||
1655 | rl->count[READ] = rl->count[WRITE] = 0; | 1655 | rl->count[READ] = rl->count[WRITE] = 0; |
1656 | rl->starved[READ] = rl->starved[WRITE] = 0; | 1656 | rl->starved[READ] = rl->starved[WRITE] = 0; |
1657 | rl->elvpriv = 0; | 1657 | rl->elvpriv = 0; |
1658 | init_waitqueue_head(&rl->wait[READ]); | 1658 | init_waitqueue_head(&rl->wait[READ]); |
1659 | init_waitqueue_head(&rl->wait[WRITE]); | 1659 | init_waitqueue_head(&rl->wait[WRITE]); |
1660 | 1660 | ||
1661 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, | 1661 | rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, |
1662 | mempool_free_slab, request_cachep, q->node); | 1662 | mempool_free_slab, request_cachep, q->node); |
1663 | 1663 | ||
1664 | if (!rl->rq_pool) | 1664 | if (!rl->rq_pool) |
1665 | return -ENOMEM; | 1665 | return -ENOMEM; |
1666 | 1666 | ||
1667 | return 0; | 1667 | return 0; |
1668 | } | 1668 | } |
1669 | 1669 | ||
1670 | static int __make_request(request_queue_t *, struct bio *); | 1670 | static int __make_request(request_queue_t *, struct bio *); |
1671 | 1671 | ||
1672 | request_queue_t *blk_alloc_queue(gfp_t gfp_mask) | 1672 | request_queue_t *blk_alloc_queue(gfp_t gfp_mask) |
1673 | { | 1673 | { |
1674 | return blk_alloc_queue_node(gfp_mask, -1); | 1674 | return blk_alloc_queue_node(gfp_mask, -1); |
1675 | } | 1675 | } |
1676 | EXPORT_SYMBOL(blk_alloc_queue); | 1676 | EXPORT_SYMBOL(blk_alloc_queue); |
1677 | 1677 | ||
1678 | request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) | 1678 | request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
1679 | { | 1679 | { |
1680 | request_queue_t *q; | 1680 | request_queue_t *q; |
1681 | 1681 | ||
1682 | q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id); | 1682 | q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id); |
1683 | if (!q) | 1683 | if (!q) |
1684 | return NULL; | 1684 | return NULL; |
1685 | 1685 | ||
1686 | memset(q, 0, sizeof(*q)); | 1686 | memset(q, 0, sizeof(*q)); |
1687 | init_timer(&q->unplug_timer); | 1687 | init_timer(&q->unplug_timer); |
1688 | atomic_set(&q->refcnt, 1); | 1688 | atomic_set(&q->refcnt, 1); |
1689 | 1689 | ||
1690 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; | 1690 | q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; |
1691 | q->backing_dev_info.unplug_io_data = q; | 1691 | q->backing_dev_info.unplug_io_data = q; |
1692 | 1692 | ||
1693 | return q; | 1693 | return q; |
1694 | } | 1694 | } |
1695 | EXPORT_SYMBOL(blk_alloc_queue_node); | 1695 | EXPORT_SYMBOL(blk_alloc_queue_node); |
1696 | 1696 | ||
1697 | /** | 1697 | /** |
1698 | * blk_init_queue - prepare a request queue for use with a block device | 1698 | * blk_init_queue - prepare a request queue for use with a block device |
1699 | * @rfn: The function to be called to process requests that have been | 1699 | * @rfn: The function to be called to process requests that have been |
1700 | * placed on the queue. | 1700 | * placed on the queue. |
1701 | * @lock: Request queue spin lock | 1701 | * @lock: Request queue spin lock |
1702 | * | 1702 | * |
1703 | * Description: | 1703 | * Description: |
1704 | * If a block device wishes to use the standard request handling procedures, | 1704 | * If a block device wishes to use the standard request handling procedures, |
1705 | * which sorts requests and coalesces adjacent requests, then it must | 1705 | * which sorts requests and coalesces adjacent requests, then it must |
1706 | * call blk_init_queue(). The function @rfn will be called when there | 1706 | * call blk_init_queue(). The function @rfn will be called when there |
1707 | * are requests on the queue that need to be processed. If the device | 1707 | * are requests on the queue that need to be processed. If the device |
1708 | * supports plugging, then @rfn may not be called immediately when requests | 1708 | * supports plugging, then @rfn may not be called immediately when requests |
1709 | * are available on the queue, but may be called at some time later instead. | 1709 | * are available on the queue, but may be called at some time later instead. |
1710 | * Plugged queues are generally unplugged when a buffer belonging to one | 1710 | * Plugged queues are generally unplugged when a buffer belonging to one |
1711 | * of the requests on the queue is needed, or due to memory pressure. | 1711 | * of the requests on the queue is needed, or due to memory pressure. |
1712 | * | 1712 | * |
1713 | * @rfn is not required, or even expected, to remove all requests off the | 1713 | * @rfn is not required, or even expected, to remove all requests off the |
1714 | * queue, but only as many as it can handle at a time. If it does leave | 1714 | * queue, but only as many as it can handle at a time. If it does leave |
1715 | * requests on the queue, it is responsible for arranging that the requests | 1715 | * requests on the queue, it is responsible for arranging that the requests |
1716 | * get dealt with eventually. | 1716 | * get dealt with eventually. |
1717 | * | 1717 | * |
1718 | * The queue spin lock must be held while manipulating the requests on the | 1718 | * The queue spin lock must be held while manipulating the requests on the |
1719 | * request queue. | 1719 | * request queue. |
1720 | * | 1720 | * |
1721 | * Function returns a pointer to the initialized request queue, or NULL if | 1721 | * Function returns a pointer to the initialized request queue, or NULL if |
1722 | * it didn't succeed. | 1722 | * it didn't succeed. |
1723 | * | 1723 | * |
1724 | * Note: | 1724 | * Note: |
1725 | * blk_init_queue() must be paired with a blk_cleanup_queue() call | 1725 | * blk_init_queue() must be paired with a blk_cleanup_queue() call |
1726 | * when the block device is deactivated (such as at module unload). | 1726 | * when the block device is deactivated (such as at module unload). |
1727 | **/ | 1727 | **/ |
1728 | 1728 | ||
1729 | request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) | 1729 | request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) |
1730 | { | 1730 | { |
1731 | return blk_init_queue_node(rfn, lock, -1); | 1731 | return blk_init_queue_node(rfn, lock, -1); |
1732 | } | 1732 | } |
1733 | EXPORT_SYMBOL(blk_init_queue); | 1733 | EXPORT_SYMBOL(blk_init_queue); |
1734 | 1734 | ||
1735 | request_queue_t * | 1735 | request_queue_t * |
1736 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) | 1736 | blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) |
1737 | { | 1737 | { |
1738 | request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id); | 1738 | request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id); |
1739 | 1739 | ||
1740 | if (!q) | 1740 | if (!q) |
1741 | return NULL; | 1741 | return NULL; |
1742 | 1742 | ||
1743 | q->node = node_id; | 1743 | q->node = node_id; |
1744 | if (blk_init_free_list(q)) | 1744 | if (blk_init_free_list(q)) |
1745 | goto out_init; | 1745 | goto out_init; |
1746 | 1746 | ||
1747 | /* | 1747 | /* |
1748 | * if caller didn't supply a lock, they get per-queue locking with | 1748 | * if caller didn't supply a lock, they get per-queue locking with |
1749 | * our embedded lock | 1749 | * our embedded lock |
1750 | */ | 1750 | */ |
1751 | if (!lock) { | 1751 | if (!lock) { |
1752 | spin_lock_init(&q->__queue_lock); | 1752 | spin_lock_init(&q->__queue_lock); |
1753 | lock = &q->__queue_lock; | 1753 | lock = &q->__queue_lock; |
1754 | } | 1754 | } |
1755 | 1755 | ||
1756 | q->request_fn = rfn; | 1756 | q->request_fn = rfn; |
1757 | q->back_merge_fn = ll_back_merge_fn; | 1757 | q->back_merge_fn = ll_back_merge_fn; |
1758 | q->front_merge_fn = ll_front_merge_fn; | 1758 | q->front_merge_fn = ll_front_merge_fn; |
1759 | q->merge_requests_fn = ll_merge_requests_fn; | 1759 | q->merge_requests_fn = ll_merge_requests_fn; |
1760 | q->prep_rq_fn = NULL; | 1760 | q->prep_rq_fn = NULL; |
1761 | q->unplug_fn = generic_unplug_device; | 1761 | q->unplug_fn = generic_unplug_device; |
1762 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); | 1762 | q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); |
1763 | q->queue_lock = lock; | 1763 | q->queue_lock = lock; |
1764 | 1764 | ||
1765 | blk_queue_segment_boundary(q, 0xffffffff); | 1765 | blk_queue_segment_boundary(q, 0xffffffff); |
1766 | 1766 | ||
1767 | blk_queue_make_request(q, __make_request); | 1767 | blk_queue_make_request(q, __make_request); |
1768 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); | 1768 | blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); |
1769 | 1769 | ||
1770 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); | 1770 | blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); |
1771 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); | 1771 | blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); |
1772 | 1772 | ||
1773 | /* | 1773 | /* |
1774 | * all done | 1774 | * all done |
1775 | */ | 1775 | */ |
1776 | if (!elevator_init(q, NULL)) { | 1776 | if (!elevator_init(q, NULL)) { |
1777 | blk_queue_congestion_threshold(q); | 1777 | blk_queue_congestion_threshold(q); |
1778 | return q; | 1778 | return q; |
1779 | } | 1779 | } |
1780 | 1780 | ||
1781 | blk_cleanup_queue(q); | 1781 | blk_cleanup_queue(q); |
1782 | out_init: | 1782 | out_init: |
1783 | kmem_cache_free(requestq_cachep, q); | 1783 | kmem_cache_free(requestq_cachep, q); |
1784 | return NULL; | 1784 | return NULL; |
1785 | } | 1785 | } |
1786 | EXPORT_SYMBOL(blk_init_queue_node); | 1786 | EXPORT_SYMBOL(blk_init_queue_node); |
1787 | 1787 | ||
1788 | int blk_get_queue(request_queue_t *q) | 1788 | int blk_get_queue(request_queue_t *q) |
1789 | { | 1789 | { |
1790 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { | 1790 | if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { |
1791 | atomic_inc(&q->refcnt); | 1791 | atomic_inc(&q->refcnt); |
1792 | return 0; | 1792 | return 0; |
1793 | } | 1793 | } |
1794 | 1794 | ||
1795 | return 1; | 1795 | return 1; |
1796 | } | 1796 | } |
1797 | 1797 | ||
1798 | EXPORT_SYMBOL(blk_get_queue); | 1798 | EXPORT_SYMBOL(blk_get_queue); |
1799 | 1799 | ||
1800 | static inline void blk_free_request(request_queue_t *q, struct request *rq) | 1800 | static inline void blk_free_request(request_queue_t *q, struct request *rq) |
1801 | { | 1801 | { |
1802 | if (rq->flags & REQ_ELVPRIV) | 1802 | if (rq->flags & REQ_ELVPRIV) |
1803 | elv_put_request(q, rq); | 1803 | elv_put_request(q, rq); |
1804 | mempool_free(rq, q->rq.rq_pool); | 1804 | mempool_free(rq, q->rq.rq_pool); |
1805 | } | 1805 | } |
1806 | 1806 | ||
1807 | static inline struct request * | 1807 | static inline struct request * |
1808 | blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, | 1808 | blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, |
1809 | int priv, gfp_t gfp_mask) | 1809 | int priv, gfp_t gfp_mask) |
1810 | { | 1810 | { |
1811 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); | 1811 | struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); |
1812 | 1812 | ||
1813 | if (!rq) | 1813 | if (!rq) |
1814 | return NULL; | 1814 | return NULL; |
1815 | 1815 | ||
1816 | /* | 1816 | /* |
1817 | * first three bits are identical in rq->flags and bio->bi_rw, | 1817 | * first three bits are identical in rq->flags and bio->bi_rw, |
1818 | * see bio.h and blkdev.h | 1818 | * see bio.h and blkdev.h |
1819 | */ | 1819 | */ |
1820 | rq->flags = rw; | 1820 | rq->flags = rw; |
1821 | 1821 | ||
1822 | if (priv) { | 1822 | if (priv) { |
1823 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) { | 1823 | if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) { |
1824 | mempool_free(rq, q->rq.rq_pool); | 1824 | mempool_free(rq, q->rq.rq_pool); |
1825 | return NULL; | 1825 | return NULL; |
1826 | } | 1826 | } |
1827 | rq->flags |= REQ_ELVPRIV; | 1827 | rq->flags |= REQ_ELVPRIV; |
1828 | } | 1828 | } |
1829 | 1829 | ||
1830 | return rq; | 1830 | return rq; |
1831 | } | 1831 | } |
1832 | 1832 | ||
1833 | /* | 1833 | /* |
1834 | * ioc_batching returns true if the ioc is a valid batching request and | 1834 | * ioc_batching returns true if the ioc is a valid batching request and |
1835 | * should be given priority access to a request. | 1835 | * should be given priority access to a request. |
1836 | */ | 1836 | */ |
1837 | static inline int ioc_batching(request_queue_t *q, struct io_context *ioc) | 1837 | static inline int ioc_batching(request_queue_t *q, struct io_context *ioc) |
1838 | { | 1838 | { |
1839 | if (!ioc) | 1839 | if (!ioc) |
1840 | return 0; | 1840 | return 0; |
1841 | 1841 | ||
1842 | /* | 1842 | /* |
1843 | * Make sure the process is able to allocate at least 1 request | 1843 | * Make sure the process is able to allocate at least 1 request |
1844 | * even if the batch times out, otherwise we could theoretically | 1844 | * even if the batch times out, otherwise we could theoretically |
1845 | * lose wakeups. | 1845 | * lose wakeups. |
1846 | */ | 1846 | */ |
1847 | return ioc->nr_batch_requests == q->nr_batching || | 1847 | return ioc->nr_batch_requests == q->nr_batching || |
1848 | (ioc->nr_batch_requests > 0 | 1848 | (ioc->nr_batch_requests > 0 |
1849 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); | 1849 | && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); |
1850 | } | 1850 | } |
1851 | 1851 | ||
1852 | /* | 1852 | /* |
1853 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This | 1853 | * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This |
1854 | * will cause the process to be a "batcher" on all queues in the system. This | 1854 | * will cause the process to be a "batcher" on all queues in the system. This |
1855 | * is the behaviour we want though - once it gets a wakeup it should be given | 1855 | * is the behaviour we want though - once it gets a wakeup it should be given |
1856 | * a nice run. | 1856 | * a nice run. |
1857 | */ | 1857 | */ |
1858 | static void ioc_set_batching(request_queue_t *q, struct io_context *ioc) | 1858 | static void ioc_set_batching(request_queue_t *q, struct io_context *ioc) |
1859 | { | 1859 | { |
1860 | if (!ioc || ioc_batching(q, ioc)) | 1860 | if (!ioc || ioc_batching(q, ioc)) |
1861 | return; | 1861 | return; |
1862 | 1862 | ||
1863 | ioc->nr_batch_requests = q->nr_batching; | 1863 | ioc->nr_batch_requests = q->nr_batching; |
1864 | ioc->last_waited = jiffies; | 1864 | ioc->last_waited = jiffies; |
1865 | } | 1865 | } |
1866 | 1866 | ||
1867 | static void __freed_request(request_queue_t *q, int rw) | 1867 | static void __freed_request(request_queue_t *q, int rw) |
1868 | { | 1868 | { |
1869 | struct request_list *rl = &q->rq; | 1869 | struct request_list *rl = &q->rq; |
1870 | 1870 | ||
1871 | if (rl->count[rw] < queue_congestion_off_threshold(q)) | 1871 | if (rl->count[rw] < queue_congestion_off_threshold(q)) |
1872 | clear_queue_congested(q, rw); | 1872 | clear_queue_congested(q, rw); |
1873 | 1873 | ||
1874 | if (rl->count[rw] + 1 <= q->nr_requests) { | 1874 | if (rl->count[rw] + 1 <= q->nr_requests) { |
1875 | if (waitqueue_active(&rl->wait[rw])) | 1875 | if (waitqueue_active(&rl->wait[rw])) |
1876 | wake_up(&rl->wait[rw]); | 1876 | wake_up(&rl->wait[rw]); |
1877 | 1877 | ||
1878 | blk_clear_queue_full(q, rw); | 1878 | blk_clear_queue_full(q, rw); |
1879 | } | 1879 | } |
1880 | } | 1880 | } |
1881 | 1881 | ||
1882 | /* | 1882 | /* |
1883 | * A request has just been released. Account for it, update the full and | 1883 | * A request has just been released. Account for it, update the full and |
1884 | * congestion status, wake up any waiters. Called under q->queue_lock. | 1884 | * congestion status, wake up any waiters. Called under q->queue_lock. |
1885 | */ | 1885 | */ |
1886 | static void freed_request(request_queue_t *q, int rw, int priv) | 1886 | static void freed_request(request_queue_t *q, int rw, int priv) |
1887 | { | 1887 | { |
1888 | struct request_list *rl = &q->rq; | 1888 | struct request_list *rl = &q->rq; |
1889 | 1889 | ||
1890 | rl->count[rw]--; | 1890 | rl->count[rw]--; |
1891 | if (priv) | 1891 | if (priv) |
1892 | rl->elvpriv--; | 1892 | rl->elvpriv--; |
1893 | 1893 | ||
1894 | __freed_request(q, rw); | 1894 | __freed_request(q, rw); |
1895 | 1895 | ||
1896 | if (unlikely(rl->starved[rw ^ 1])) | 1896 | if (unlikely(rl->starved[rw ^ 1])) |
1897 | __freed_request(q, rw ^ 1); | 1897 | __freed_request(q, rw ^ 1); |
1898 | } | 1898 | } |
1899 | 1899 | ||
1900 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) | 1900 | #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) |
1901 | /* | 1901 | /* |
1902 | * Get a free request, queue_lock must be held. | 1902 | * Get a free request, queue_lock must be held. |
1903 | * Returns NULL on failure, with queue_lock held. | 1903 | * Returns NULL on failure, with queue_lock held. |
1904 | * Returns !NULL on success, with queue_lock *not held*. | 1904 | * Returns !NULL on success, with queue_lock *not held*. |
1905 | */ | 1905 | */ |
1906 | static struct request *get_request(request_queue_t *q, int rw, struct bio *bio, | 1906 | static struct request *get_request(request_queue_t *q, int rw, struct bio *bio, |
1907 | gfp_t gfp_mask) | 1907 | gfp_t gfp_mask) |
1908 | { | 1908 | { |
1909 | struct request *rq = NULL; | 1909 | struct request *rq = NULL; |
1910 | struct request_list *rl = &q->rq; | 1910 | struct request_list *rl = &q->rq; |
1911 | struct io_context *ioc = current_io_context(GFP_ATOMIC); | 1911 | struct io_context *ioc = NULL; |
1912 | int priv; | 1912 | int may_queue, priv; |
1913 | 1913 | ||
1914 | if (rl->count[rw]+1 >= q->nr_requests) { | 1914 | may_queue = elv_may_queue(q, rw, bio); |
1915 | /* | 1915 | if (may_queue == ELV_MQUEUE_NO) |
1916 | * The queue will fill after this allocation, so set it as | 1916 | goto rq_starved; |
1917 | * full, and mark this process as "batching". This process | 1917 | |
1918 | * will be allowed to complete a batch of requests, others | 1918 | if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { |
1919 | * will be blocked. | 1919 | if (rl->count[rw]+1 >= q->nr_requests) { |
1920 | */ | 1920 | ioc = current_io_context(GFP_ATOMIC); |
1921 | if (!blk_queue_full(q, rw)) { | 1921 | /* |
1922 | ioc_set_batching(q, ioc); | 1922 | * The queue will fill after this allocation, so set |
1923 | blk_set_queue_full(q, rw); | 1923 | * it as full, and mark this process as "batching". |
1924 | * This process will be allowed to complete a batch of | ||
1925 | * requests, others will be blocked. | ||
1926 | */ | ||
1927 | if (!blk_queue_full(q, rw)) { | ||
1928 | ioc_set_batching(q, ioc); | ||
1929 | blk_set_queue_full(q, rw); | ||
1930 | } else { | ||
1931 | if (may_queue != ELV_MQUEUE_MUST | ||
1932 | && !ioc_batching(q, ioc)) { | ||
1933 | /* | ||
1934 | * The queue is full and the allocating | ||
1935 | * process is not a "batcher", and not | ||
1936 | * exempted by the IO scheduler | ||
1937 | */ | ||
1938 | goto out; | ||
1939 | } | ||
1940 | } | ||
1924 | } | 1941 | } |
1942 | set_queue_congested(q, rw); | ||
1925 | } | 1943 | } |
1926 | 1944 | ||
1927 | switch (elv_may_queue(q, rw, bio)) { | ||
1928 | case ELV_MQUEUE_NO: | ||
1929 | goto rq_starved; | ||
1930 | case ELV_MQUEUE_MAY: | ||
1931 | break; | ||
1932 | case ELV_MQUEUE_MUST: | ||
1933 | goto get_rq; | ||
1934 | } | ||
1935 | |||
1936 | if (blk_queue_full(q, rw) && !ioc_batching(q, ioc)) { | ||
1937 | /* | ||
1938 | * The queue is full and the allocating process is not a | ||
1939 | * "batcher", and not exempted by the IO scheduler | ||
1940 | */ | ||
1941 | goto out; | ||
1942 | } | ||
1943 | |||
1944 | get_rq: | ||
1945 | /* | 1945 | /* |
1946 | * Only allow batching queuers to allocate up to 50% over the defined | 1946 | * Only allow batching queuers to allocate up to 50% over the defined |
1947 | * limit of requests, otherwise we could have thousands of requests | 1947 | * limit of requests, otherwise we could have thousands of requests |
1948 | * allocated with any setting of ->nr_requests | 1948 | * allocated with any setting of ->nr_requests |
1949 | */ | 1949 | */ |
1950 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) | 1950 | if (rl->count[rw] >= (3 * q->nr_requests / 2)) |
1951 | goto out; | 1951 | goto out; |
1952 | 1952 | ||
1953 | rl->count[rw]++; | 1953 | rl->count[rw]++; |
1954 | rl->starved[rw] = 0; | 1954 | rl->starved[rw] = 0; |
1955 | if (rl->count[rw] >= queue_congestion_on_threshold(q)) | ||
1956 | set_queue_congested(q, rw); | ||
1957 | 1955 | ||
1958 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); | 1956 | priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1959 | if (priv) | 1957 | if (priv) |
1960 | rl->elvpriv++; | 1958 | rl->elvpriv++; |
1961 | 1959 | ||
1962 | spin_unlock_irq(q->queue_lock); | 1960 | spin_unlock_irq(q->queue_lock); |
1963 | 1961 | ||
1964 | rq = blk_alloc_request(q, rw, bio, priv, gfp_mask); | 1962 | rq = blk_alloc_request(q, rw, bio, priv, gfp_mask); |
1965 | if (!rq) { | 1963 | if (unlikely(!rq)) { |
1966 | /* | 1964 | /* |
1967 | * Allocation failed presumably due to memory. Undo anything | 1965 | * Allocation failed presumably due to memory. Undo anything |
1968 | * we might have messed up. | 1966 | * we might have messed up. |
1969 | * | 1967 | * |
1970 | * Allocating task should really be put onto the front of the | 1968 | * Allocating task should really be put onto the front of the |
1971 | * wait queue, but this is pretty rare. | 1969 | * wait queue, but this is pretty rare. |
1972 | */ | 1970 | */ |
1973 | spin_lock_irq(q->queue_lock); | 1971 | spin_lock_irq(q->queue_lock); |
1974 | freed_request(q, rw, priv); | 1972 | freed_request(q, rw, priv); |
1975 | 1973 | ||
1976 | /* | 1974 | /* |
1977 | * in the very unlikely event that allocation failed and no | 1975 | * in the very unlikely event that allocation failed and no |
1978 | * requests for this direction was pending, mark us starved | 1976 | * requests for this direction was pending, mark us starved |
1979 | * so that freeing of a request in the other direction will | 1977 | * so that freeing of a request in the other direction will |
1980 | * notice us. another possible fix would be to split the | 1978 | * notice us. another possible fix would be to split the |
1981 | * rq mempool into READ and WRITE | 1979 | * rq mempool into READ and WRITE |
1982 | */ | 1980 | */ |
1983 | rq_starved: | 1981 | rq_starved: |
1984 | if (unlikely(rl->count[rw] == 0)) | 1982 | if (unlikely(rl->count[rw] == 0)) |
1985 | rl->starved[rw] = 1; | 1983 | rl->starved[rw] = 1; |
1986 | 1984 | ||
1987 | goto out; | 1985 | goto out; |
1988 | } | 1986 | } |
1989 | 1987 | ||
1988 | /* | ||
1989 | * ioc may be NULL here, and ioc_batching will be false. That's | ||
1990 | * OK, if the queue is under the request limit then requests need | ||
1991 | * not count toward the nr_batch_requests limit. There will always | ||
1992 | * be some limit enforced by BLK_BATCH_TIME. | ||
1993 | */ | ||
1990 | if (ioc_batching(q, ioc)) | 1994 | if (ioc_batching(q, ioc)) |
1991 | ioc->nr_batch_requests--; | 1995 | ioc->nr_batch_requests--; |
1992 | 1996 | ||
1993 | rq_init(q, rq); | 1997 | rq_init(q, rq); |
1994 | rq->rl = rl; | 1998 | rq->rl = rl; |
1995 | out: | 1999 | out: |
1996 | return rq; | 2000 | return rq; |
1997 | } | 2001 | } |
1998 | 2002 | ||
1999 | /* | 2003 | /* |
2000 | * No available requests for this queue, unplug the device and wait for some | 2004 | * No available requests for this queue, unplug the device and wait for some |
2001 | * requests to become available. | 2005 | * requests to become available. |
2002 | * | 2006 | * |
2003 | * Called with q->queue_lock held, and returns with it unlocked. | 2007 | * Called with q->queue_lock held, and returns with it unlocked. |
2004 | */ | 2008 | */ |
2005 | static struct request *get_request_wait(request_queue_t *q, int rw, | 2009 | static struct request *get_request_wait(request_queue_t *q, int rw, |
2006 | struct bio *bio) | 2010 | struct bio *bio) |
2007 | { | 2011 | { |
2008 | struct request *rq; | 2012 | struct request *rq; |
2009 | 2013 | ||
2010 | rq = get_request(q, rw, bio, GFP_NOIO); | 2014 | rq = get_request(q, rw, bio, GFP_NOIO); |
2011 | while (!rq) { | 2015 | while (!rq) { |
2012 | DEFINE_WAIT(wait); | 2016 | DEFINE_WAIT(wait); |
2013 | struct request_list *rl = &q->rq; | 2017 | struct request_list *rl = &q->rq; |
2014 | 2018 | ||
2015 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, | 2019 | prepare_to_wait_exclusive(&rl->wait[rw], &wait, |
2016 | TASK_UNINTERRUPTIBLE); | 2020 | TASK_UNINTERRUPTIBLE); |
2017 | 2021 | ||
2018 | rq = get_request(q, rw, bio, GFP_NOIO); | 2022 | rq = get_request(q, rw, bio, GFP_NOIO); |
2019 | 2023 | ||
2020 | if (!rq) { | 2024 | if (!rq) { |
2021 | struct io_context *ioc; | 2025 | struct io_context *ioc; |
2022 | 2026 | ||
2023 | __generic_unplug_device(q); | 2027 | __generic_unplug_device(q); |
2024 | spin_unlock_irq(q->queue_lock); | 2028 | spin_unlock_irq(q->queue_lock); |
2025 | io_schedule(); | 2029 | io_schedule(); |
2026 | 2030 | ||
2027 | /* | 2031 | /* |
2028 | * After sleeping, we become a "batching" process and | 2032 | * After sleeping, we become a "batching" process and |
2029 | * will be able to allocate at least one request, and | 2033 | * will be able to allocate at least one request, and |
2030 | * up to a big batch of them for a small period time. | 2034 | * up to a big batch of them for a small period time. |
2031 | * See ioc_batching, ioc_set_batching | 2035 | * See ioc_batching, ioc_set_batching |
2032 | */ | 2036 | */ |
2033 | ioc = current_io_context(GFP_NOIO); | 2037 | ioc = current_io_context(GFP_NOIO); |
2034 | ioc_set_batching(q, ioc); | 2038 | ioc_set_batching(q, ioc); |
2035 | 2039 | ||
2036 | spin_lock_irq(q->queue_lock); | 2040 | spin_lock_irq(q->queue_lock); |
2037 | } | 2041 | } |
2038 | finish_wait(&rl->wait[rw], &wait); | 2042 | finish_wait(&rl->wait[rw], &wait); |
2039 | } | 2043 | } |
2040 | 2044 | ||
2041 | return rq; | 2045 | return rq; |
2042 | } | 2046 | } |
2043 | 2047 | ||
2044 | struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask) | 2048 | struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask) |
2045 | { | 2049 | { |
2046 | struct request *rq; | 2050 | struct request *rq; |
2047 | 2051 | ||
2048 | BUG_ON(rw != READ && rw != WRITE); | 2052 | BUG_ON(rw != READ && rw != WRITE); |
2049 | 2053 | ||
2050 | spin_lock_irq(q->queue_lock); | 2054 | spin_lock_irq(q->queue_lock); |
2051 | if (gfp_mask & __GFP_WAIT) { | 2055 | if (gfp_mask & __GFP_WAIT) { |
2052 | rq = get_request_wait(q, rw, NULL); | 2056 | rq = get_request_wait(q, rw, NULL); |
2053 | } else { | 2057 | } else { |
2054 | rq = get_request(q, rw, NULL, gfp_mask); | 2058 | rq = get_request(q, rw, NULL, gfp_mask); |
2055 | if (!rq) | 2059 | if (!rq) |
2056 | spin_unlock_irq(q->queue_lock); | 2060 | spin_unlock_irq(q->queue_lock); |
2057 | } | 2061 | } |
2058 | /* q->queue_lock is unlocked at this point */ | 2062 | /* q->queue_lock is unlocked at this point */ |
2059 | 2063 | ||
2060 | return rq; | 2064 | return rq; |
2061 | } | 2065 | } |
2062 | EXPORT_SYMBOL(blk_get_request); | 2066 | EXPORT_SYMBOL(blk_get_request); |
2063 | 2067 | ||
2064 | /** | 2068 | /** |
2065 | * blk_requeue_request - put a request back on queue | 2069 | * blk_requeue_request - put a request back on queue |
2066 | * @q: request queue where request should be inserted | 2070 | * @q: request queue where request should be inserted |
2067 | * @rq: request to be inserted | 2071 | * @rq: request to be inserted |
2068 | * | 2072 | * |
2069 | * Description: | 2073 | * Description: |
2070 | * Drivers often keep queueing requests until the hardware cannot accept | 2074 | * Drivers often keep queueing requests until the hardware cannot accept |
2071 | * more, when that condition happens we need to put the request back | 2075 | * more, when that condition happens we need to put the request back |
2072 | * on the queue. Must be called with queue lock held. | 2076 | * on the queue. Must be called with queue lock held. |
2073 | */ | 2077 | */ |
2074 | void blk_requeue_request(request_queue_t *q, struct request *rq) | 2078 | void blk_requeue_request(request_queue_t *q, struct request *rq) |
2075 | { | 2079 | { |
2076 | if (blk_rq_tagged(rq)) | 2080 | if (blk_rq_tagged(rq)) |
2077 | blk_queue_end_tag(q, rq); | 2081 | blk_queue_end_tag(q, rq); |
2078 | 2082 | ||
2079 | elv_requeue_request(q, rq); | 2083 | elv_requeue_request(q, rq); |
2080 | } | 2084 | } |
2081 | 2085 | ||
2082 | EXPORT_SYMBOL(blk_requeue_request); | 2086 | EXPORT_SYMBOL(blk_requeue_request); |
2083 | 2087 | ||
2084 | /** | 2088 | /** |
2085 | * blk_insert_request - insert a special request in to a request queue | 2089 | * blk_insert_request - insert a special request in to a request queue |
2086 | * @q: request queue where request should be inserted | 2090 | * @q: request queue where request should be inserted |
2087 | * @rq: request to be inserted | 2091 | * @rq: request to be inserted |
2088 | * @at_head: insert request at head or tail of queue | 2092 | * @at_head: insert request at head or tail of queue |
2089 | * @data: private data | 2093 | * @data: private data |
2090 | * | 2094 | * |
2091 | * Description: | 2095 | * Description: |
2092 | * Many block devices need to execute commands asynchronously, so they don't | 2096 | * Many block devices need to execute commands asynchronously, so they don't |
2093 | * block the whole kernel from preemption during request execution. This is | 2097 | * block the whole kernel from preemption during request execution. This is |
2094 | * accomplished normally by inserting aritficial requests tagged as | 2098 | * accomplished normally by inserting aritficial requests tagged as |
2095 | * REQ_SPECIAL in to the corresponding request queue, and letting them be | 2099 | * REQ_SPECIAL in to the corresponding request queue, and letting them be |
2096 | * scheduled for actual execution by the request queue. | 2100 | * scheduled for actual execution by the request queue. |
2097 | * | 2101 | * |
2098 | * We have the option of inserting the head or the tail of the queue. | 2102 | * We have the option of inserting the head or the tail of the queue. |
2099 | * Typically we use the tail for new ioctls and so forth. We use the head | 2103 | * Typically we use the tail for new ioctls and so forth. We use the head |
2100 | * of the queue for things like a QUEUE_FULL message from a device, or a | 2104 | * of the queue for things like a QUEUE_FULL message from a device, or a |
2101 | * host that is unable to accept a particular command. | 2105 | * host that is unable to accept a particular command. |
2102 | */ | 2106 | */ |
2103 | void blk_insert_request(request_queue_t *q, struct request *rq, | 2107 | void blk_insert_request(request_queue_t *q, struct request *rq, |
2104 | int at_head, void *data) | 2108 | int at_head, void *data) |
2105 | { | 2109 | { |
2106 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | 2110 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
2107 | unsigned long flags; | 2111 | unsigned long flags; |
2108 | 2112 | ||
2109 | /* | 2113 | /* |
2110 | * tell I/O scheduler that this isn't a regular read/write (ie it | 2114 | * tell I/O scheduler that this isn't a regular read/write (ie it |
2111 | * must not attempt merges on this) and that it acts as a soft | 2115 | * must not attempt merges on this) and that it acts as a soft |
2112 | * barrier | 2116 | * barrier |
2113 | */ | 2117 | */ |
2114 | rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER; | 2118 | rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER; |
2115 | 2119 | ||
2116 | rq->special = data; | 2120 | rq->special = data; |
2117 | 2121 | ||
2118 | spin_lock_irqsave(q->queue_lock, flags); | 2122 | spin_lock_irqsave(q->queue_lock, flags); |
2119 | 2123 | ||
2120 | /* | 2124 | /* |
2121 | * If command is tagged, release the tag | 2125 | * If command is tagged, release the tag |
2122 | */ | 2126 | */ |
2123 | if (blk_rq_tagged(rq)) | 2127 | if (blk_rq_tagged(rq)) |
2124 | blk_queue_end_tag(q, rq); | 2128 | blk_queue_end_tag(q, rq); |
2125 | 2129 | ||
2126 | drive_stat_acct(rq, rq->nr_sectors, 1); | 2130 | drive_stat_acct(rq, rq->nr_sectors, 1); |
2127 | __elv_add_request(q, rq, where, 0); | 2131 | __elv_add_request(q, rq, where, 0); |
2128 | 2132 | ||
2129 | if (blk_queue_plugged(q)) | 2133 | if (blk_queue_plugged(q)) |
2130 | __generic_unplug_device(q); | 2134 | __generic_unplug_device(q); |
2131 | else | 2135 | else |
2132 | q->request_fn(q); | 2136 | q->request_fn(q); |
2133 | spin_unlock_irqrestore(q->queue_lock, flags); | 2137 | spin_unlock_irqrestore(q->queue_lock, flags); |
2134 | } | 2138 | } |
2135 | 2139 | ||
2136 | EXPORT_SYMBOL(blk_insert_request); | 2140 | EXPORT_SYMBOL(blk_insert_request); |
2137 | 2141 | ||
2138 | /** | 2142 | /** |
2139 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage | 2143 | * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage |
2140 | * @q: request queue where request should be inserted | 2144 | * @q: request queue where request should be inserted |
2141 | * @rq: request structure to fill | 2145 | * @rq: request structure to fill |
2142 | * @ubuf: the user buffer | 2146 | * @ubuf: the user buffer |
2143 | * @len: length of user data | 2147 | * @len: length of user data |
2144 | * | 2148 | * |
2145 | * Description: | 2149 | * Description: |
2146 | * Data will be mapped directly for zero copy io, if possible. Otherwise | 2150 | * Data will be mapped directly for zero copy io, if possible. Otherwise |
2147 | * a kernel bounce buffer is used. | 2151 | * a kernel bounce buffer is used. |
2148 | * | 2152 | * |
2149 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | 2153 | * A matching blk_rq_unmap_user() must be issued at the end of io, while |
2150 | * still in process context. | 2154 | * still in process context. |
2151 | * | 2155 | * |
2152 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | 2156 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() |
2153 | * before being submitted to the device, as pages mapped may be out of | 2157 | * before being submitted to the device, as pages mapped may be out of |
2154 | * reach. It's the callers responsibility to make sure this happens. The | 2158 | * reach. It's the callers responsibility to make sure this happens. The |
2155 | * original bio must be passed back in to blk_rq_unmap_user() for proper | 2159 | * original bio must be passed back in to blk_rq_unmap_user() for proper |
2156 | * unmapping. | 2160 | * unmapping. |
2157 | */ | 2161 | */ |
2158 | int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf, | 2162 | int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf, |
2159 | unsigned int len) | 2163 | unsigned int len) |
2160 | { | 2164 | { |
2161 | unsigned long uaddr; | 2165 | unsigned long uaddr; |
2162 | struct bio *bio; | 2166 | struct bio *bio; |
2163 | int reading; | 2167 | int reading; |
2164 | 2168 | ||
2165 | if (len > (q->max_hw_sectors << 9)) | 2169 | if (len > (q->max_hw_sectors << 9)) |
2166 | return -EINVAL; | 2170 | return -EINVAL; |
2167 | if (!len || !ubuf) | 2171 | if (!len || !ubuf) |
2168 | return -EINVAL; | 2172 | return -EINVAL; |
2169 | 2173 | ||
2170 | reading = rq_data_dir(rq) == READ; | 2174 | reading = rq_data_dir(rq) == READ; |
2171 | 2175 | ||
2172 | /* | 2176 | /* |
2173 | * if alignment requirement is satisfied, map in user pages for | 2177 | * if alignment requirement is satisfied, map in user pages for |
2174 | * direct dma. else, set up kernel bounce buffers | 2178 | * direct dma. else, set up kernel bounce buffers |
2175 | */ | 2179 | */ |
2176 | uaddr = (unsigned long) ubuf; | 2180 | uaddr = (unsigned long) ubuf; |
2177 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) | 2181 | if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) |
2178 | bio = bio_map_user(q, NULL, uaddr, len, reading); | 2182 | bio = bio_map_user(q, NULL, uaddr, len, reading); |
2179 | else | 2183 | else |
2180 | bio = bio_copy_user(q, uaddr, len, reading); | 2184 | bio = bio_copy_user(q, uaddr, len, reading); |
2181 | 2185 | ||
2182 | if (!IS_ERR(bio)) { | 2186 | if (!IS_ERR(bio)) { |
2183 | rq->bio = rq->biotail = bio; | 2187 | rq->bio = rq->biotail = bio; |
2184 | blk_rq_bio_prep(q, rq, bio); | 2188 | blk_rq_bio_prep(q, rq, bio); |
2185 | 2189 | ||
2186 | rq->buffer = rq->data = NULL; | 2190 | rq->buffer = rq->data = NULL; |
2187 | rq->data_len = len; | 2191 | rq->data_len = len; |
2188 | return 0; | 2192 | return 0; |
2189 | } | 2193 | } |
2190 | 2194 | ||
2191 | /* | 2195 | /* |
2192 | * bio is the err-ptr | 2196 | * bio is the err-ptr |
2193 | */ | 2197 | */ |
2194 | return PTR_ERR(bio); | 2198 | return PTR_ERR(bio); |
2195 | } | 2199 | } |
2196 | 2200 | ||
2197 | EXPORT_SYMBOL(blk_rq_map_user); | 2201 | EXPORT_SYMBOL(blk_rq_map_user); |
2198 | 2202 | ||
2199 | /** | 2203 | /** |
2200 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage | 2204 | * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage |
2201 | * @q: request queue where request should be inserted | 2205 | * @q: request queue where request should be inserted |
2202 | * @rq: request to map data to | 2206 | * @rq: request to map data to |
2203 | * @iov: pointer to the iovec | 2207 | * @iov: pointer to the iovec |
2204 | * @iov_count: number of elements in the iovec | 2208 | * @iov_count: number of elements in the iovec |
2205 | * | 2209 | * |
2206 | * Description: | 2210 | * Description: |
2207 | * Data will be mapped directly for zero copy io, if possible. Otherwise | 2211 | * Data will be mapped directly for zero copy io, if possible. Otherwise |
2208 | * a kernel bounce buffer is used. | 2212 | * a kernel bounce buffer is used. |
2209 | * | 2213 | * |
2210 | * A matching blk_rq_unmap_user() must be issued at the end of io, while | 2214 | * A matching blk_rq_unmap_user() must be issued at the end of io, while |
2211 | * still in process context. | 2215 | * still in process context. |
2212 | * | 2216 | * |
2213 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() | 2217 | * Note: The mapped bio may need to be bounced through blk_queue_bounce() |
2214 | * before being submitted to the device, as pages mapped may be out of | 2218 | * before being submitted to the device, as pages mapped may be out of |
2215 | * reach. It's the callers responsibility to make sure this happens. The | 2219 | * reach. It's the callers responsibility to make sure this happens. The |
2216 | * original bio must be passed back in to blk_rq_unmap_user() for proper | 2220 | * original bio must be passed back in to blk_rq_unmap_user() for proper |
2217 | * unmapping. | 2221 | * unmapping. |
2218 | */ | 2222 | */ |
2219 | int blk_rq_map_user_iov(request_queue_t *q, struct request *rq, | 2223 | int blk_rq_map_user_iov(request_queue_t *q, struct request *rq, |
2220 | struct sg_iovec *iov, int iov_count) | 2224 | struct sg_iovec *iov, int iov_count) |
2221 | { | 2225 | { |
2222 | struct bio *bio; | 2226 | struct bio *bio; |
2223 | 2227 | ||
2224 | if (!iov || iov_count <= 0) | 2228 | if (!iov || iov_count <= 0) |
2225 | return -EINVAL; | 2229 | return -EINVAL; |
2226 | 2230 | ||
2227 | /* we don't allow misaligned data like bio_map_user() does. If the | 2231 | /* we don't allow misaligned data like bio_map_user() does. If the |
2228 | * user is using sg, they're expected to know the alignment constraints | 2232 | * user is using sg, they're expected to know the alignment constraints |
2229 | * and respect them accordingly */ | 2233 | * and respect them accordingly */ |
2230 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); | 2234 | bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); |
2231 | if (IS_ERR(bio)) | 2235 | if (IS_ERR(bio)) |
2232 | return PTR_ERR(bio); | 2236 | return PTR_ERR(bio); |
2233 | 2237 | ||
2234 | rq->bio = rq->biotail = bio; | 2238 | rq->bio = rq->biotail = bio; |
2235 | blk_rq_bio_prep(q, rq, bio); | 2239 | blk_rq_bio_prep(q, rq, bio); |
2236 | rq->buffer = rq->data = NULL; | 2240 | rq->buffer = rq->data = NULL; |
2237 | rq->data_len = bio->bi_size; | 2241 | rq->data_len = bio->bi_size; |
2238 | return 0; | 2242 | return 0; |
2239 | } | 2243 | } |
2240 | 2244 | ||
2241 | EXPORT_SYMBOL(blk_rq_map_user_iov); | 2245 | EXPORT_SYMBOL(blk_rq_map_user_iov); |
2242 | 2246 | ||
2243 | /** | 2247 | /** |
2244 | * blk_rq_unmap_user - unmap a request with user data | 2248 | * blk_rq_unmap_user - unmap a request with user data |
2245 | * @bio: bio to be unmapped | 2249 | * @bio: bio to be unmapped |
2246 | * @ulen: length of user buffer | 2250 | * @ulen: length of user buffer |
2247 | * | 2251 | * |
2248 | * Description: | 2252 | * Description: |
2249 | * Unmap a bio previously mapped by blk_rq_map_user(). | 2253 | * Unmap a bio previously mapped by blk_rq_map_user(). |
2250 | */ | 2254 | */ |
2251 | int blk_rq_unmap_user(struct bio *bio, unsigned int ulen) | 2255 | int blk_rq_unmap_user(struct bio *bio, unsigned int ulen) |
2252 | { | 2256 | { |
2253 | int ret = 0; | 2257 | int ret = 0; |
2254 | 2258 | ||
2255 | if (bio) { | 2259 | if (bio) { |
2256 | if (bio_flagged(bio, BIO_USER_MAPPED)) | 2260 | if (bio_flagged(bio, BIO_USER_MAPPED)) |
2257 | bio_unmap_user(bio); | 2261 | bio_unmap_user(bio); |
2258 | else | 2262 | else |
2259 | ret = bio_uncopy_user(bio); | 2263 | ret = bio_uncopy_user(bio); |
2260 | } | 2264 | } |
2261 | 2265 | ||
2262 | return 0; | 2266 | return 0; |
2263 | } | 2267 | } |
2264 | 2268 | ||
2265 | EXPORT_SYMBOL(blk_rq_unmap_user); | 2269 | EXPORT_SYMBOL(blk_rq_unmap_user); |
2266 | 2270 | ||
2267 | /** | 2271 | /** |
2268 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage | 2272 | * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage |
2269 | * @q: request queue where request should be inserted | 2273 | * @q: request queue where request should be inserted |
2270 | * @rq: request to fill | 2274 | * @rq: request to fill |
2271 | * @kbuf: the kernel buffer | 2275 | * @kbuf: the kernel buffer |
2272 | * @len: length of user data | 2276 | * @len: length of user data |
2273 | * @gfp_mask: memory allocation flags | 2277 | * @gfp_mask: memory allocation flags |
2274 | */ | 2278 | */ |
2275 | int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf, | 2279 | int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf, |
2276 | unsigned int len, gfp_t gfp_mask) | 2280 | unsigned int len, gfp_t gfp_mask) |
2277 | { | 2281 | { |
2278 | struct bio *bio; | 2282 | struct bio *bio; |
2279 | 2283 | ||
2280 | if (len > (q->max_hw_sectors << 9)) | 2284 | if (len > (q->max_hw_sectors << 9)) |
2281 | return -EINVAL; | 2285 | return -EINVAL; |
2282 | if (!len || !kbuf) | 2286 | if (!len || !kbuf) |
2283 | return -EINVAL; | 2287 | return -EINVAL; |
2284 | 2288 | ||
2285 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | 2289 | bio = bio_map_kern(q, kbuf, len, gfp_mask); |
2286 | if (IS_ERR(bio)) | 2290 | if (IS_ERR(bio)) |
2287 | return PTR_ERR(bio); | 2291 | return PTR_ERR(bio); |
2288 | 2292 | ||
2289 | if (rq_data_dir(rq) == WRITE) | 2293 | if (rq_data_dir(rq) == WRITE) |
2290 | bio->bi_rw |= (1 << BIO_RW); | 2294 | bio->bi_rw |= (1 << BIO_RW); |
2291 | 2295 | ||
2292 | rq->bio = rq->biotail = bio; | 2296 | rq->bio = rq->biotail = bio; |
2293 | blk_rq_bio_prep(q, rq, bio); | 2297 | blk_rq_bio_prep(q, rq, bio); |
2294 | 2298 | ||
2295 | rq->buffer = rq->data = NULL; | 2299 | rq->buffer = rq->data = NULL; |
2296 | rq->data_len = len; | 2300 | rq->data_len = len; |
2297 | return 0; | 2301 | return 0; |
2298 | } | 2302 | } |
2299 | 2303 | ||
2300 | EXPORT_SYMBOL(blk_rq_map_kern); | 2304 | EXPORT_SYMBOL(blk_rq_map_kern); |
2301 | 2305 | ||
2302 | /** | 2306 | /** |
2303 | * blk_execute_rq_nowait - insert a request into queue for execution | 2307 | * blk_execute_rq_nowait - insert a request into queue for execution |
2304 | * @q: queue to insert the request in | 2308 | * @q: queue to insert the request in |
2305 | * @bd_disk: matching gendisk | 2309 | * @bd_disk: matching gendisk |
2306 | * @rq: request to insert | 2310 | * @rq: request to insert |
2307 | * @at_head: insert request at head or tail of queue | 2311 | * @at_head: insert request at head or tail of queue |
2308 | * @done: I/O completion handler | 2312 | * @done: I/O completion handler |
2309 | * | 2313 | * |
2310 | * Description: | 2314 | * Description: |
2311 | * Insert a fully prepared request at the back of the io scheduler queue | 2315 | * Insert a fully prepared request at the back of the io scheduler queue |
2312 | * for execution. Don't wait for completion. | 2316 | * for execution. Don't wait for completion. |
2313 | */ | 2317 | */ |
2314 | void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk, | 2318 | void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk, |
2315 | struct request *rq, int at_head, | 2319 | struct request *rq, int at_head, |
2316 | void (*done)(struct request *)) | 2320 | void (*done)(struct request *)) |
2317 | { | 2321 | { |
2318 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; | 2322 | int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; |
2319 | 2323 | ||
2320 | rq->rq_disk = bd_disk; | 2324 | rq->rq_disk = bd_disk; |
2321 | rq->flags |= REQ_NOMERGE; | 2325 | rq->flags |= REQ_NOMERGE; |
2322 | rq->end_io = done; | 2326 | rq->end_io = done; |
2323 | elv_add_request(q, rq, where, 1); | 2327 | elv_add_request(q, rq, where, 1); |
2324 | generic_unplug_device(q); | 2328 | generic_unplug_device(q); |
2325 | } | 2329 | } |
2326 | 2330 | ||
2327 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); | 2331 | EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); |
2328 | 2332 | ||
2329 | /** | 2333 | /** |
2330 | * blk_execute_rq - insert a request into queue for execution | 2334 | * blk_execute_rq - insert a request into queue for execution |
2331 | * @q: queue to insert the request in | 2335 | * @q: queue to insert the request in |
2332 | * @bd_disk: matching gendisk | 2336 | * @bd_disk: matching gendisk |
2333 | * @rq: request to insert | 2337 | * @rq: request to insert |
2334 | * @at_head: insert request at head or tail of queue | 2338 | * @at_head: insert request at head or tail of queue |
2335 | * | 2339 | * |
2336 | * Description: | 2340 | * Description: |
2337 | * Insert a fully prepared request at the back of the io scheduler queue | 2341 | * Insert a fully prepared request at the back of the io scheduler queue |
2338 | * for execution and wait for completion. | 2342 | * for execution and wait for completion. |
2339 | */ | 2343 | */ |
2340 | int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk, | 2344 | int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk, |
2341 | struct request *rq, int at_head) | 2345 | struct request *rq, int at_head) |
2342 | { | 2346 | { |
2343 | DECLARE_COMPLETION(wait); | 2347 | DECLARE_COMPLETION(wait); |
2344 | char sense[SCSI_SENSE_BUFFERSIZE]; | 2348 | char sense[SCSI_SENSE_BUFFERSIZE]; |
2345 | int err = 0; | 2349 | int err = 0; |
2346 | 2350 | ||
2347 | /* | 2351 | /* |
2348 | * we need an extra reference to the request, so we can look at | 2352 | * we need an extra reference to the request, so we can look at |
2349 | * it after io completion | 2353 | * it after io completion |
2350 | */ | 2354 | */ |
2351 | rq->ref_count++; | 2355 | rq->ref_count++; |
2352 | 2356 | ||
2353 | if (!rq->sense) { | 2357 | if (!rq->sense) { |
2354 | memset(sense, 0, sizeof(sense)); | 2358 | memset(sense, 0, sizeof(sense)); |
2355 | rq->sense = sense; | 2359 | rq->sense = sense; |
2356 | rq->sense_len = 0; | 2360 | rq->sense_len = 0; |
2357 | } | 2361 | } |
2358 | 2362 | ||
2359 | rq->waiting = &wait; | 2363 | rq->waiting = &wait; |
2360 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); | 2364 | blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); |
2361 | wait_for_completion(&wait); | 2365 | wait_for_completion(&wait); |
2362 | rq->waiting = NULL; | 2366 | rq->waiting = NULL; |
2363 | 2367 | ||
2364 | if (rq->errors) | 2368 | if (rq->errors) |
2365 | err = -EIO; | 2369 | err = -EIO; |
2366 | 2370 | ||
2367 | return err; | 2371 | return err; |
2368 | } | 2372 | } |
2369 | 2373 | ||
2370 | EXPORT_SYMBOL(blk_execute_rq); | 2374 | EXPORT_SYMBOL(blk_execute_rq); |
2371 | 2375 | ||
2372 | /** | 2376 | /** |
2373 | * blkdev_issue_flush - queue a flush | 2377 | * blkdev_issue_flush - queue a flush |
2374 | * @bdev: blockdev to issue flush for | 2378 | * @bdev: blockdev to issue flush for |
2375 | * @error_sector: error sector | 2379 | * @error_sector: error sector |
2376 | * | 2380 | * |
2377 | * Description: | 2381 | * Description: |
2378 | * Issue a flush for the block device in question. Caller can supply | 2382 | * Issue a flush for the block device in question. Caller can supply |
2379 | * room for storing the error offset in case of a flush error, if they | 2383 | * room for storing the error offset in case of a flush error, if they |
2380 | * wish to. Caller must run wait_for_completion() on its own. | 2384 | * wish to. Caller must run wait_for_completion() on its own. |
2381 | */ | 2385 | */ |
2382 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) | 2386 | int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) |
2383 | { | 2387 | { |
2384 | request_queue_t *q; | 2388 | request_queue_t *q; |
2385 | 2389 | ||
2386 | if (bdev->bd_disk == NULL) | 2390 | if (bdev->bd_disk == NULL) |
2387 | return -ENXIO; | 2391 | return -ENXIO; |
2388 | 2392 | ||
2389 | q = bdev_get_queue(bdev); | 2393 | q = bdev_get_queue(bdev); |
2390 | if (!q) | 2394 | if (!q) |
2391 | return -ENXIO; | 2395 | return -ENXIO; |
2392 | if (!q->issue_flush_fn) | 2396 | if (!q->issue_flush_fn) |
2393 | return -EOPNOTSUPP; | 2397 | return -EOPNOTSUPP; |
2394 | 2398 | ||
2395 | return q->issue_flush_fn(q, bdev->bd_disk, error_sector); | 2399 | return q->issue_flush_fn(q, bdev->bd_disk, error_sector); |
2396 | } | 2400 | } |
2397 | 2401 | ||
2398 | EXPORT_SYMBOL(blkdev_issue_flush); | 2402 | EXPORT_SYMBOL(blkdev_issue_flush); |
2399 | 2403 | ||
2400 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) | 2404 | static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) |
2401 | { | 2405 | { |
2402 | int rw = rq_data_dir(rq); | 2406 | int rw = rq_data_dir(rq); |
2403 | 2407 | ||
2404 | if (!blk_fs_request(rq) || !rq->rq_disk) | 2408 | if (!blk_fs_request(rq) || !rq->rq_disk) |
2405 | return; | 2409 | return; |
2406 | 2410 | ||
2407 | if (!new_io) { | 2411 | if (!new_io) { |
2408 | __disk_stat_inc(rq->rq_disk, merges[rw]); | 2412 | __disk_stat_inc(rq->rq_disk, merges[rw]); |
2409 | } else { | 2413 | } else { |
2410 | disk_round_stats(rq->rq_disk); | 2414 | disk_round_stats(rq->rq_disk); |
2411 | rq->rq_disk->in_flight++; | 2415 | rq->rq_disk->in_flight++; |
2412 | } | 2416 | } |
2413 | } | 2417 | } |
2414 | 2418 | ||
2415 | /* | 2419 | /* |
2416 | * add-request adds a request to the linked list. | 2420 | * add-request adds a request to the linked list. |
2417 | * queue lock is held and interrupts disabled, as we muck with the | 2421 | * queue lock is held and interrupts disabled, as we muck with the |
2418 | * request queue list. | 2422 | * request queue list. |
2419 | */ | 2423 | */ |
2420 | static inline void add_request(request_queue_t * q, struct request * req) | 2424 | static inline void add_request(request_queue_t * q, struct request * req) |
2421 | { | 2425 | { |
2422 | drive_stat_acct(req, req->nr_sectors, 1); | 2426 | drive_stat_acct(req, req->nr_sectors, 1); |
2423 | 2427 | ||
2424 | if (q->activity_fn) | 2428 | if (q->activity_fn) |
2425 | q->activity_fn(q->activity_data, rq_data_dir(req)); | 2429 | q->activity_fn(q->activity_data, rq_data_dir(req)); |
2426 | 2430 | ||
2427 | /* | 2431 | /* |
2428 | * elevator indicated where it wants this request to be | 2432 | * elevator indicated where it wants this request to be |
2429 | * inserted at elevator_merge time | 2433 | * inserted at elevator_merge time |
2430 | */ | 2434 | */ |
2431 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); | 2435 | __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); |
2432 | } | 2436 | } |
2433 | 2437 | ||
2434 | /* | 2438 | /* |
2435 | * disk_round_stats() - Round off the performance stats on a struct | 2439 | * disk_round_stats() - Round off the performance stats on a struct |
2436 | * disk_stats. | 2440 | * disk_stats. |
2437 | * | 2441 | * |
2438 | * The average IO queue length and utilisation statistics are maintained | 2442 | * The average IO queue length and utilisation statistics are maintained |
2439 | * by observing the current state of the queue length and the amount of | 2443 | * by observing the current state of the queue length and the amount of |
2440 | * time it has been in this state for. | 2444 | * time it has been in this state for. |
2441 | * | 2445 | * |
2442 | * Normally, that accounting is done on IO completion, but that can result | 2446 | * Normally, that accounting is done on IO completion, but that can result |
2443 | * in more than a second's worth of IO being accounted for within any one | 2447 | * in more than a second's worth of IO being accounted for within any one |
2444 | * second, leading to >100% utilisation. To deal with that, we call this | 2448 | * second, leading to >100% utilisation. To deal with that, we call this |
2445 | * function to do a round-off before returning the results when reading | 2449 | * function to do a round-off before returning the results when reading |
2446 | * /proc/diskstats. This accounts immediately for all queue usage up to | 2450 | * /proc/diskstats. This accounts immediately for all queue usage up to |
2447 | * the current jiffies and restarts the counters again. | 2451 | * the current jiffies and restarts the counters again. |
2448 | */ | 2452 | */ |
2449 | void disk_round_stats(struct gendisk *disk) | 2453 | void disk_round_stats(struct gendisk *disk) |
2450 | { | 2454 | { |
2451 | unsigned long now = jiffies; | 2455 | unsigned long now = jiffies; |
2452 | 2456 | ||
2453 | if (now == disk->stamp) | 2457 | if (now == disk->stamp) |
2454 | return; | 2458 | return; |
2455 | 2459 | ||
2456 | if (disk->in_flight) { | 2460 | if (disk->in_flight) { |
2457 | __disk_stat_add(disk, time_in_queue, | 2461 | __disk_stat_add(disk, time_in_queue, |
2458 | disk->in_flight * (now - disk->stamp)); | 2462 | disk->in_flight * (now - disk->stamp)); |
2459 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); | 2463 | __disk_stat_add(disk, io_ticks, (now - disk->stamp)); |
2460 | } | 2464 | } |
2461 | disk->stamp = now; | 2465 | disk->stamp = now; |
2462 | } | 2466 | } |
2463 | 2467 | ||
2464 | /* | 2468 | /* |
2465 | * queue lock must be held | 2469 | * queue lock must be held |
2466 | */ | 2470 | */ |
2467 | void __blk_put_request(request_queue_t *q, struct request *req) | 2471 | void __blk_put_request(request_queue_t *q, struct request *req) |
2468 | { | 2472 | { |
2469 | struct request_list *rl = req->rl; | 2473 | struct request_list *rl = req->rl; |
2470 | 2474 | ||
2471 | if (unlikely(!q)) | 2475 | if (unlikely(!q)) |
2472 | return; | 2476 | return; |
2473 | if (unlikely(--req->ref_count)) | 2477 | if (unlikely(--req->ref_count)) |
2474 | return; | 2478 | return; |
2475 | 2479 | ||
2476 | elv_completed_request(q, req); | 2480 | elv_completed_request(q, req); |
2477 | 2481 | ||
2478 | req->rq_status = RQ_INACTIVE; | 2482 | req->rq_status = RQ_INACTIVE; |
2479 | req->rl = NULL; | 2483 | req->rl = NULL; |
2480 | 2484 | ||
2481 | /* | 2485 | /* |
2482 | * Request may not have originated from ll_rw_blk. if not, | 2486 | * Request may not have originated from ll_rw_blk. if not, |
2483 | * it didn't come out of our reserved rq pools | 2487 | * it didn't come out of our reserved rq pools |
2484 | */ | 2488 | */ |
2485 | if (rl) { | 2489 | if (rl) { |
2486 | int rw = rq_data_dir(req); | 2490 | int rw = rq_data_dir(req); |
2487 | int priv = req->flags & REQ_ELVPRIV; | 2491 | int priv = req->flags & REQ_ELVPRIV; |
2488 | 2492 | ||
2489 | BUG_ON(!list_empty(&req->queuelist)); | 2493 | BUG_ON(!list_empty(&req->queuelist)); |
2490 | 2494 | ||
2491 | blk_free_request(q, req); | 2495 | blk_free_request(q, req); |
2492 | freed_request(q, rw, priv); | 2496 | freed_request(q, rw, priv); |
2493 | } | 2497 | } |
2494 | } | 2498 | } |
2495 | 2499 | ||
2496 | EXPORT_SYMBOL_GPL(__blk_put_request); | 2500 | EXPORT_SYMBOL_GPL(__blk_put_request); |
2497 | 2501 | ||
2498 | void blk_put_request(struct request *req) | 2502 | void blk_put_request(struct request *req) |
2499 | { | 2503 | { |
2500 | unsigned long flags; | 2504 | unsigned long flags; |
2501 | request_queue_t *q = req->q; | 2505 | request_queue_t *q = req->q; |
2502 | 2506 | ||
2503 | /* | 2507 | /* |
2504 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the | 2508 | * Gee, IDE calls in w/ NULL q. Fix IDE and remove the |
2505 | * following if (q) test. | 2509 | * following if (q) test. |
2506 | */ | 2510 | */ |
2507 | if (q) { | 2511 | if (q) { |
2508 | spin_lock_irqsave(q->queue_lock, flags); | 2512 | spin_lock_irqsave(q->queue_lock, flags); |
2509 | __blk_put_request(q, req); | 2513 | __blk_put_request(q, req); |
2510 | spin_unlock_irqrestore(q->queue_lock, flags); | 2514 | spin_unlock_irqrestore(q->queue_lock, flags); |
2511 | } | 2515 | } |
2512 | } | 2516 | } |
2513 | 2517 | ||
2514 | EXPORT_SYMBOL(blk_put_request); | 2518 | EXPORT_SYMBOL(blk_put_request); |
2515 | 2519 | ||
2516 | /** | 2520 | /** |
2517 | * blk_end_sync_rq - executes a completion event on a request | 2521 | * blk_end_sync_rq - executes a completion event on a request |
2518 | * @rq: request to complete | 2522 | * @rq: request to complete |
2519 | */ | 2523 | */ |
2520 | void blk_end_sync_rq(struct request *rq) | 2524 | void blk_end_sync_rq(struct request *rq) |
2521 | { | 2525 | { |
2522 | struct completion *waiting = rq->waiting; | 2526 | struct completion *waiting = rq->waiting; |
2523 | 2527 | ||
2524 | rq->waiting = NULL; | 2528 | rq->waiting = NULL; |
2525 | __blk_put_request(rq->q, rq); | 2529 | __blk_put_request(rq->q, rq); |
2526 | 2530 | ||
2527 | /* | 2531 | /* |
2528 | * complete last, if this is a stack request the process (and thus | 2532 | * complete last, if this is a stack request the process (and thus |
2529 | * the rq pointer) could be invalid right after this complete() | 2533 | * the rq pointer) could be invalid right after this complete() |
2530 | */ | 2534 | */ |
2531 | complete(waiting); | 2535 | complete(waiting); |
2532 | } | 2536 | } |
2533 | EXPORT_SYMBOL(blk_end_sync_rq); | 2537 | EXPORT_SYMBOL(blk_end_sync_rq); |
2534 | 2538 | ||
2535 | /** | 2539 | /** |
2536 | * blk_congestion_wait - wait for a queue to become uncongested | 2540 | * blk_congestion_wait - wait for a queue to become uncongested |
2537 | * @rw: READ or WRITE | 2541 | * @rw: READ or WRITE |
2538 | * @timeout: timeout in jiffies | 2542 | * @timeout: timeout in jiffies |
2539 | * | 2543 | * |
2540 | * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion. | 2544 | * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion. |
2541 | * If no queues are congested then just wait for the next request to be | 2545 | * If no queues are congested then just wait for the next request to be |
2542 | * returned. | 2546 | * returned. |
2543 | */ | 2547 | */ |
2544 | long blk_congestion_wait(int rw, long timeout) | 2548 | long blk_congestion_wait(int rw, long timeout) |
2545 | { | 2549 | { |
2546 | long ret; | 2550 | long ret; |
2547 | DEFINE_WAIT(wait); | 2551 | DEFINE_WAIT(wait); |
2548 | wait_queue_head_t *wqh = &congestion_wqh[rw]; | 2552 | wait_queue_head_t *wqh = &congestion_wqh[rw]; |
2549 | 2553 | ||
2550 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | 2554 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); |
2551 | ret = io_schedule_timeout(timeout); | 2555 | ret = io_schedule_timeout(timeout); |
2552 | finish_wait(wqh, &wait); | 2556 | finish_wait(wqh, &wait); |
2553 | return ret; | 2557 | return ret; |
2554 | } | 2558 | } |
2555 | 2559 | ||
2556 | EXPORT_SYMBOL(blk_congestion_wait); | 2560 | EXPORT_SYMBOL(blk_congestion_wait); |
2557 | 2561 | ||
2558 | /* | 2562 | /* |
2559 | * Has to be called with the request spinlock acquired | 2563 | * Has to be called with the request spinlock acquired |
2560 | */ | 2564 | */ |
2561 | static int attempt_merge(request_queue_t *q, struct request *req, | 2565 | static int attempt_merge(request_queue_t *q, struct request *req, |
2562 | struct request *next) | 2566 | struct request *next) |
2563 | { | 2567 | { |
2564 | if (!rq_mergeable(req) || !rq_mergeable(next)) | 2568 | if (!rq_mergeable(req) || !rq_mergeable(next)) |
2565 | return 0; | 2569 | return 0; |
2566 | 2570 | ||
2567 | /* | 2571 | /* |
2568 | * not contigious | 2572 | * not contigious |
2569 | */ | 2573 | */ |
2570 | if (req->sector + req->nr_sectors != next->sector) | 2574 | if (req->sector + req->nr_sectors != next->sector) |
2571 | return 0; | 2575 | return 0; |
2572 | 2576 | ||
2573 | if (rq_data_dir(req) != rq_data_dir(next) | 2577 | if (rq_data_dir(req) != rq_data_dir(next) |
2574 | || req->rq_disk != next->rq_disk | 2578 | || req->rq_disk != next->rq_disk |
2575 | || next->waiting || next->special) | 2579 | || next->waiting || next->special) |
2576 | return 0; | 2580 | return 0; |
2577 | 2581 | ||
2578 | /* | 2582 | /* |
2579 | * If we are allowed to merge, then append bio list | 2583 | * If we are allowed to merge, then append bio list |
2580 | * from next to rq and release next. merge_requests_fn | 2584 | * from next to rq and release next. merge_requests_fn |
2581 | * will have updated segment counts, update sector | 2585 | * will have updated segment counts, update sector |
2582 | * counts here. | 2586 | * counts here. |
2583 | */ | 2587 | */ |
2584 | if (!q->merge_requests_fn(q, req, next)) | 2588 | if (!q->merge_requests_fn(q, req, next)) |
2585 | return 0; | 2589 | return 0; |
2586 | 2590 | ||
2587 | /* | 2591 | /* |
2588 | * At this point we have either done a back merge | 2592 | * At this point we have either done a back merge |
2589 | * or front merge. We need the smaller start_time of | 2593 | * or front merge. We need the smaller start_time of |
2590 | * the merged requests to be the current request | 2594 | * the merged requests to be the current request |
2591 | * for accounting purposes. | 2595 | * for accounting purposes. |
2592 | */ | 2596 | */ |
2593 | if (time_after(req->start_time, next->start_time)) | 2597 | if (time_after(req->start_time, next->start_time)) |
2594 | req->start_time = next->start_time; | 2598 | req->start_time = next->start_time; |
2595 | 2599 | ||
2596 | req->biotail->bi_next = next->bio; | 2600 | req->biotail->bi_next = next->bio; |
2597 | req->biotail = next->biotail; | 2601 | req->biotail = next->biotail; |
2598 | 2602 | ||
2599 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; | 2603 | req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; |
2600 | 2604 | ||
2601 | elv_merge_requests(q, req, next); | 2605 | elv_merge_requests(q, req, next); |
2602 | 2606 | ||
2603 | if (req->rq_disk) { | 2607 | if (req->rq_disk) { |
2604 | disk_round_stats(req->rq_disk); | 2608 | disk_round_stats(req->rq_disk); |
2605 | req->rq_disk->in_flight--; | 2609 | req->rq_disk->in_flight--; |
2606 | } | 2610 | } |
2607 | 2611 | ||
2608 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); | 2612 | req->ioprio = ioprio_best(req->ioprio, next->ioprio); |
2609 | 2613 | ||
2610 | __blk_put_request(q, next); | 2614 | __blk_put_request(q, next); |
2611 | return 1; | 2615 | return 1; |
2612 | } | 2616 | } |
2613 | 2617 | ||
2614 | static inline int attempt_back_merge(request_queue_t *q, struct request *rq) | 2618 | static inline int attempt_back_merge(request_queue_t *q, struct request *rq) |
2615 | { | 2619 | { |
2616 | struct request *next = elv_latter_request(q, rq); | 2620 | struct request *next = elv_latter_request(q, rq); |
2617 | 2621 | ||
2618 | if (next) | 2622 | if (next) |
2619 | return attempt_merge(q, rq, next); | 2623 | return attempt_merge(q, rq, next); |
2620 | 2624 | ||
2621 | return 0; | 2625 | return 0; |
2622 | } | 2626 | } |
2623 | 2627 | ||
2624 | static inline int attempt_front_merge(request_queue_t *q, struct request *rq) | 2628 | static inline int attempt_front_merge(request_queue_t *q, struct request *rq) |
2625 | { | 2629 | { |
2626 | struct request *prev = elv_former_request(q, rq); | 2630 | struct request *prev = elv_former_request(q, rq); |
2627 | 2631 | ||
2628 | if (prev) | 2632 | if (prev) |
2629 | return attempt_merge(q, prev, rq); | 2633 | return attempt_merge(q, prev, rq); |
2630 | 2634 | ||
2631 | return 0; | 2635 | return 0; |
2632 | } | 2636 | } |
2633 | 2637 | ||
2634 | /** | 2638 | /** |
2635 | * blk_attempt_remerge - attempt to remerge active head with next request | 2639 | * blk_attempt_remerge - attempt to remerge active head with next request |
2636 | * @q: The &request_queue_t belonging to the device | 2640 | * @q: The &request_queue_t belonging to the device |
2637 | * @rq: The head request (usually) | 2641 | * @rq: The head request (usually) |
2638 | * | 2642 | * |
2639 | * Description: | 2643 | * Description: |
2640 | * For head-active devices, the queue can easily be unplugged so quickly | 2644 | * For head-active devices, the queue can easily be unplugged so quickly |
2641 | * that proper merging is not done on the front request. This may hurt | 2645 | * that proper merging is not done on the front request. This may hurt |
2642 | * performance greatly for some devices. The block layer cannot safely | 2646 | * performance greatly for some devices. The block layer cannot safely |
2643 | * do merging on that first request for these queues, but the driver can | 2647 | * do merging on that first request for these queues, but the driver can |
2644 | * call this function and make it happen any way. Only the driver knows | 2648 | * call this function and make it happen any way. Only the driver knows |
2645 | * when it is safe to do so. | 2649 | * when it is safe to do so. |
2646 | **/ | 2650 | **/ |
2647 | void blk_attempt_remerge(request_queue_t *q, struct request *rq) | 2651 | void blk_attempt_remerge(request_queue_t *q, struct request *rq) |
2648 | { | 2652 | { |
2649 | unsigned long flags; | 2653 | unsigned long flags; |
2650 | 2654 | ||
2651 | spin_lock_irqsave(q->queue_lock, flags); | 2655 | spin_lock_irqsave(q->queue_lock, flags); |
2652 | attempt_back_merge(q, rq); | 2656 | attempt_back_merge(q, rq); |
2653 | spin_unlock_irqrestore(q->queue_lock, flags); | 2657 | spin_unlock_irqrestore(q->queue_lock, flags); |
2654 | } | 2658 | } |
2655 | 2659 | ||
2656 | EXPORT_SYMBOL(blk_attempt_remerge); | 2660 | EXPORT_SYMBOL(blk_attempt_remerge); |
2657 | 2661 | ||
2658 | static int __make_request(request_queue_t *q, struct bio *bio) | 2662 | static int __make_request(request_queue_t *q, struct bio *bio) |
2659 | { | 2663 | { |
2660 | struct request *req; | 2664 | struct request *req; |
2661 | int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync; | 2665 | int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync; |
2662 | unsigned short prio; | 2666 | unsigned short prio; |
2663 | sector_t sector; | 2667 | sector_t sector; |
2664 | 2668 | ||
2665 | sector = bio->bi_sector; | 2669 | sector = bio->bi_sector; |
2666 | nr_sectors = bio_sectors(bio); | 2670 | nr_sectors = bio_sectors(bio); |
2667 | cur_nr_sectors = bio_cur_sectors(bio); | 2671 | cur_nr_sectors = bio_cur_sectors(bio); |
2668 | prio = bio_prio(bio); | 2672 | prio = bio_prio(bio); |
2669 | 2673 | ||
2670 | rw = bio_data_dir(bio); | 2674 | rw = bio_data_dir(bio); |
2671 | sync = bio_sync(bio); | 2675 | sync = bio_sync(bio); |
2672 | 2676 | ||
2673 | /* | 2677 | /* |
2674 | * low level driver can indicate that it wants pages above a | 2678 | * low level driver can indicate that it wants pages above a |
2675 | * certain limit bounced to low memory (ie for highmem, or even | 2679 | * certain limit bounced to low memory (ie for highmem, or even |
2676 | * ISA dma in theory) | 2680 | * ISA dma in theory) |
2677 | */ | 2681 | */ |
2678 | blk_queue_bounce(q, &bio); | 2682 | blk_queue_bounce(q, &bio); |
2679 | 2683 | ||
2680 | spin_lock_prefetch(q->queue_lock); | 2684 | spin_lock_prefetch(q->queue_lock); |
2681 | 2685 | ||
2682 | barrier = bio_barrier(bio); | 2686 | barrier = bio_barrier(bio); |
2683 | if (unlikely(barrier) && (q->ordered == QUEUE_ORDERED_NONE)) { | 2687 | if (unlikely(barrier) && (q->ordered == QUEUE_ORDERED_NONE)) { |
2684 | err = -EOPNOTSUPP; | 2688 | err = -EOPNOTSUPP; |
2685 | goto end_io; | 2689 | goto end_io; |
2686 | } | 2690 | } |
2687 | 2691 | ||
2688 | spin_lock_irq(q->queue_lock); | 2692 | spin_lock_irq(q->queue_lock); |
2689 | 2693 | ||
2690 | if (unlikely(barrier) || elv_queue_empty(q)) | 2694 | if (unlikely(barrier) || elv_queue_empty(q)) |
2691 | goto get_rq; | 2695 | goto get_rq; |
2692 | 2696 | ||
2693 | el_ret = elv_merge(q, &req, bio); | 2697 | el_ret = elv_merge(q, &req, bio); |
2694 | switch (el_ret) { | 2698 | switch (el_ret) { |
2695 | case ELEVATOR_BACK_MERGE: | 2699 | case ELEVATOR_BACK_MERGE: |
2696 | BUG_ON(!rq_mergeable(req)); | 2700 | BUG_ON(!rq_mergeable(req)); |
2697 | 2701 | ||
2698 | if (!q->back_merge_fn(q, req, bio)) | 2702 | if (!q->back_merge_fn(q, req, bio)) |
2699 | break; | 2703 | break; |
2700 | 2704 | ||
2701 | req->biotail->bi_next = bio; | 2705 | req->biotail->bi_next = bio; |
2702 | req->biotail = bio; | 2706 | req->biotail = bio; |
2703 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | 2707 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
2704 | req->ioprio = ioprio_best(req->ioprio, prio); | 2708 | req->ioprio = ioprio_best(req->ioprio, prio); |
2705 | drive_stat_acct(req, nr_sectors, 0); | 2709 | drive_stat_acct(req, nr_sectors, 0); |
2706 | if (!attempt_back_merge(q, req)) | 2710 | if (!attempt_back_merge(q, req)) |
2707 | elv_merged_request(q, req); | 2711 | elv_merged_request(q, req); |
2708 | goto out; | 2712 | goto out; |
2709 | 2713 | ||
2710 | case ELEVATOR_FRONT_MERGE: | 2714 | case ELEVATOR_FRONT_MERGE: |
2711 | BUG_ON(!rq_mergeable(req)); | 2715 | BUG_ON(!rq_mergeable(req)); |
2712 | 2716 | ||
2713 | if (!q->front_merge_fn(q, req, bio)) | 2717 | if (!q->front_merge_fn(q, req, bio)) |
2714 | break; | 2718 | break; |
2715 | 2719 | ||
2716 | bio->bi_next = req->bio; | 2720 | bio->bi_next = req->bio; |
2717 | req->bio = bio; | 2721 | req->bio = bio; |
2718 | 2722 | ||
2719 | /* | 2723 | /* |
2720 | * may not be valid. if the low level driver said | 2724 | * may not be valid. if the low level driver said |
2721 | * it didn't need a bounce buffer then it better | 2725 | * it didn't need a bounce buffer then it better |
2722 | * not touch req->buffer either... | 2726 | * not touch req->buffer either... |
2723 | */ | 2727 | */ |
2724 | req->buffer = bio_data(bio); | 2728 | req->buffer = bio_data(bio); |
2725 | req->current_nr_sectors = cur_nr_sectors; | 2729 | req->current_nr_sectors = cur_nr_sectors; |
2726 | req->hard_cur_sectors = cur_nr_sectors; | 2730 | req->hard_cur_sectors = cur_nr_sectors; |
2727 | req->sector = req->hard_sector = sector; | 2731 | req->sector = req->hard_sector = sector; |
2728 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; | 2732 | req->nr_sectors = req->hard_nr_sectors += nr_sectors; |
2729 | req->ioprio = ioprio_best(req->ioprio, prio); | 2733 | req->ioprio = ioprio_best(req->ioprio, prio); |
2730 | drive_stat_acct(req, nr_sectors, 0); | 2734 | drive_stat_acct(req, nr_sectors, 0); |
2731 | if (!attempt_front_merge(q, req)) | 2735 | if (!attempt_front_merge(q, req)) |
2732 | elv_merged_request(q, req); | 2736 | elv_merged_request(q, req); |
2733 | goto out; | 2737 | goto out; |
2734 | 2738 | ||
2735 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ | 2739 | /* ELV_NO_MERGE: elevator says don't/can't merge. */ |
2736 | default: | 2740 | default: |
2737 | ; | 2741 | ; |
2738 | } | 2742 | } |
2739 | 2743 | ||
2740 | get_rq: | 2744 | get_rq: |
2741 | /* | 2745 | /* |
2742 | * Grab a free request. This is might sleep but can not fail. | 2746 | * Grab a free request. This is might sleep but can not fail. |
2743 | * Returns with the queue unlocked. | 2747 | * Returns with the queue unlocked. |
2744 | */ | 2748 | */ |
2745 | req = get_request_wait(q, rw, bio); | 2749 | req = get_request_wait(q, rw, bio); |
2746 | 2750 | ||
2747 | /* | 2751 | /* |
2748 | * After dropping the lock and possibly sleeping here, our request | 2752 | * After dropping the lock and possibly sleeping here, our request |
2749 | * may now be mergeable after it had proven unmergeable (above). | 2753 | * may now be mergeable after it had proven unmergeable (above). |
2750 | * We don't worry about that case for efficiency. It won't happen | 2754 | * We don't worry about that case for efficiency. It won't happen |
2751 | * often, and the elevators are able to handle it. | 2755 | * often, and the elevators are able to handle it. |
2752 | */ | 2756 | */ |
2753 | 2757 | ||
2754 | req->flags |= REQ_CMD; | 2758 | req->flags |= REQ_CMD; |
2755 | 2759 | ||
2756 | /* | 2760 | /* |
2757 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) | 2761 | * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) |
2758 | */ | 2762 | */ |
2759 | if (bio_rw_ahead(bio) || bio_failfast(bio)) | 2763 | if (bio_rw_ahead(bio) || bio_failfast(bio)) |
2760 | req->flags |= REQ_FAILFAST; | 2764 | req->flags |= REQ_FAILFAST; |
2761 | 2765 | ||
2762 | /* | 2766 | /* |
2763 | * REQ_BARRIER implies no merging, but lets make it explicit | 2767 | * REQ_BARRIER implies no merging, but lets make it explicit |
2764 | */ | 2768 | */ |
2765 | if (unlikely(barrier)) | 2769 | if (unlikely(barrier)) |
2766 | req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE); | 2770 | req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE); |
2767 | 2771 | ||
2768 | req->errors = 0; | 2772 | req->errors = 0; |
2769 | req->hard_sector = req->sector = sector; | 2773 | req->hard_sector = req->sector = sector; |
2770 | req->hard_nr_sectors = req->nr_sectors = nr_sectors; | 2774 | req->hard_nr_sectors = req->nr_sectors = nr_sectors; |
2771 | req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors; | 2775 | req->current_nr_sectors = req->hard_cur_sectors = cur_nr_sectors; |
2772 | req->nr_phys_segments = bio_phys_segments(q, bio); | 2776 | req->nr_phys_segments = bio_phys_segments(q, bio); |
2773 | req->nr_hw_segments = bio_hw_segments(q, bio); | 2777 | req->nr_hw_segments = bio_hw_segments(q, bio); |
2774 | req->buffer = bio_data(bio); /* see ->buffer comment above */ | 2778 | req->buffer = bio_data(bio); /* see ->buffer comment above */ |
2775 | req->waiting = NULL; | 2779 | req->waiting = NULL; |
2776 | req->bio = req->biotail = bio; | 2780 | req->bio = req->biotail = bio; |
2777 | req->ioprio = prio; | 2781 | req->ioprio = prio; |
2778 | req->rq_disk = bio->bi_bdev->bd_disk; | 2782 | req->rq_disk = bio->bi_bdev->bd_disk; |
2779 | req->start_time = jiffies; | 2783 | req->start_time = jiffies; |
2780 | 2784 | ||
2781 | spin_lock_irq(q->queue_lock); | 2785 | spin_lock_irq(q->queue_lock); |
2782 | if (elv_queue_empty(q)) | 2786 | if (elv_queue_empty(q)) |
2783 | blk_plug_device(q); | 2787 | blk_plug_device(q); |
2784 | add_request(q, req); | 2788 | add_request(q, req); |
2785 | out: | 2789 | out: |
2786 | if (sync) | 2790 | if (sync) |
2787 | __generic_unplug_device(q); | 2791 | __generic_unplug_device(q); |
2788 | 2792 | ||
2789 | spin_unlock_irq(q->queue_lock); | 2793 | spin_unlock_irq(q->queue_lock); |
2790 | return 0; | 2794 | return 0; |
2791 | 2795 | ||
2792 | end_io: | 2796 | end_io: |
2793 | bio_endio(bio, nr_sectors << 9, err); | 2797 | bio_endio(bio, nr_sectors << 9, err); |
2794 | return 0; | 2798 | return 0; |
2795 | } | 2799 | } |
2796 | 2800 | ||
2797 | /* | 2801 | /* |
2798 | * If bio->bi_dev is a partition, remap the location | 2802 | * If bio->bi_dev is a partition, remap the location |
2799 | */ | 2803 | */ |
2800 | static inline void blk_partition_remap(struct bio *bio) | 2804 | static inline void blk_partition_remap(struct bio *bio) |
2801 | { | 2805 | { |
2802 | struct block_device *bdev = bio->bi_bdev; | 2806 | struct block_device *bdev = bio->bi_bdev; |
2803 | 2807 | ||
2804 | if (bdev != bdev->bd_contains) { | 2808 | if (bdev != bdev->bd_contains) { |
2805 | struct hd_struct *p = bdev->bd_part; | 2809 | struct hd_struct *p = bdev->bd_part; |
2806 | const int rw = bio_data_dir(bio); | 2810 | const int rw = bio_data_dir(bio); |
2807 | 2811 | ||
2808 | p->sectors[rw] += bio_sectors(bio); | 2812 | p->sectors[rw] += bio_sectors(bio); |
2809 | p->ios[rw]++; | 2813 | p->ios[rw]++; |
2810 | 2814 | ||
2811 | bio->bi_sector += p->start_sect; | 2815 | bio->bi_sector += p->start_sect; |
2812 | bio->bi_bdev = bdev->bd_contains; | 2816 | bio->bi_bdev = bdev->bd_contains; |
2813 | } | 2817 | } |
2814 | } | 2818 | } |
2815 | 2819 | ||
2816 | static void handle_bad_sector(struct bio *bio) | 2820 | static void handle_bad_sector(struct bio *bio) |
2817 | { | 2821 | { |
2818 | char b[BDEVNAME_SIZE]; | 2822 | char b[BDEVNAME_SIZE]; |
2819 | 2823 | ||
2820 | printk(KERN_INFO "attempt to access beyond end of device\n"); | 2824 | printk(KERN_INFO "attempt to access beyond end of device\n"); |
2821 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", | 2825 | printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", |
2822 | bdevname(bio->bi_bdev, b), | 2826 | bdevname(bio->bi_bdev, b), |
2823 | bio->bi_rw, | 2827 | bio->bi_rw, |
2824 | (unsigned long long)bio->bi_sector + bio_sectors(bio), | 2828 | (unsigned long long)bio->bi_sector + bio_sectors(bio), |
2825 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); | 2829 | (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); |
2826 | 2830 | ||
2827 | set_bit(BIO_EOF, &bio->bi_flags); | 2831 | set_bit(BIO_EOF, &bio->bi_flags); |
2828 | } | 2832 | } |
2829 | 2833 | ||
2830 | /** | 2834 | /** |
2831 | * generic_make_request: hand a buffer to its device driver for I/O | 2835 | * generic_make_request: hand a buffer to its device driver for I/O |
2832 | * @bio: The bio describing the location in memory and on the device. | 2836 | * @bio: The bio describing the location in memory and on the device. |
2833 | * | 2837 | * |
2834 | * generic_make_request() is used to make I/O requests of block | 2838 | * generic_make_request() is used to make I/O requests of block |
2835 | * devices. It is passed a &struct bio, which describes the I/O that needs | 2839 | * devices. It is passed a &struct bio, which describes the I/O that needs |
2836 | * to be done. | 2840 | * to be done. |
2837 | * | 2841 | * |
2838 | * generic_make_request() does not return any status. The | 2842 | * generic_make_request() does not return any status. The |
2839 | * success/failure status of the request, along with notification of | 2843 | * success/failure status of the request, along with notification of |
2840 | * completion, is delivered asynchronously through the bio->bi_end_io | 2844 | * completion, is delivered asynchronously through the bio->bi_end_io |
2841 | * function described (one day) else where. | 2845 | * function described (one day) else where. |
2842 | * | 2846 | * |
2843 | * The caller of generic_make_request must make sure that bi_io_vec | 2847 | * The caller of generic_make_request must make sure that bi_io_vec |
2844 | * are set to describe the memory buffer, and that bi_dev and bi_sector are | 2848 | * are set to describe the memory buffer, and that bi_dev and bi_sector are |
2845 | * set to describe the device address, and the | 2849 | * set to describe the device address, and the |
2846 | * bi_end_io and optionally bi_private are set to describe how | 2850 | * bi_end_io and optionally bi_private are set to describe how |
2847 | * completion notification should be signaled. | 2851 | * completion notification should be signaled. |
2848 | * | 2852 | * |
2849 | * generic_make_request and the drivers it calls may use bi_next if this | 2853 | * generic_make_request and the drivers it calls may use bi_next if this |
2850 | * bio happens to be merged with someone else, and may change bi_dev and | 2854 | * bio happens to be merged with someone else, and may change bi_dev and |
2851 | * bi_sector for remaps as it sees fit. So the values of these fields | 2855 | * bi_sector for remaps as it sees fit. So the values of these fields |
2852 | * should NOT be depended on after the call to generic_make_request. | 2856 | * should NOT be depended on after the call to generic_make_request. |
2853 | */ | 2857 | */ |
2854 | void generic_make_request(struct bio *bio) | 2858 | void generic_make_request(struct bio *bio) |
2855 | { | 2859 | { |
2856 | request_queue_t *q; | 2860 | request_queue_t *q; |
2857 | sector_t maxsector; | 2861 | sector_t maxsector; |
2858 | int ret, nr_sectors = bio_sectors(bio); | 2862 | int ret, nr_sectors = bio_sectors(bio); |
2859 | 2863 | ||
2860 | might_sleep(); | 2864 | might_sleep(); |
2861 | /* Test device or partition size, when known. */ | 2865 | /* Test device or partition size, when known. */ |
2862 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; | 2866 | maxsector = bio->bi_bdev->bd_inode->i_size >> 9; |
2863 | if (maxsector) { | 2867 | if (maxsector) { |
2864 | sector_t sector = bio->bi_sector; | 2868 | sector_t sector = bio->bi_sector; |
2865 | 2869 | ||
2866 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { | 2870 | if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { |
2867 | /* | 2871 | /* |
2868 | * This may well happen - the kernel calls bread() | 2872 | * This may well happen - the kernel calls bread() |
2869 | * without checking the size of the device, e.g., when | 2873 | * without checking the size of the device, e.g., when |
2870 | * mounting a device. | 2874 | * mounting a device. |
2871 | */ | 2875 | */ |
2872 | handle_bad_sector(bio); | 2876 | handle_bad_sector(bio); |
2873 | goto end_io; | 2877 | goto end_io; |
2874 | } | 2878 | } |
2875 | } | 2879 | } |
2876 | 2880 | ||
2877 | /* | 2881 | /* |
2878 | * Resolve the mapping until finished. (drivers are | 2882 | * Resolve the mapping until finished. (drivers are |
2879 | * still free to implement/resolve their own stacking | 2883 | * still free to implement/resolve their own stacking |
2880 | * by explicitly returning 0) | 2884 | * by explicitly returning 0) |
2881 | * | 2885 | * |
2882 | * NOTE: we don't repeat the blk_size check for each new device. | 2886 | * NOTE: we don't repeat the blk_size check for each new device. |
2883 | * Stacking drivers are expected to know what they are doing. | 2887 | * Stacking drivers are expected to know what they are doing. |
2884 | */ | 2888 | */ |
2885 | do { | 2889 | do { |
2886 | char b[BDEVNAME_SIZE]; | 2890 | char b[BDEVNAME_SIZE]; |
2887 | 2891 | ||
2888 | q = bdev_get_queue(bio->bi_bdev); | 2892 | q = bdev_get_queue(bio->bi_bdev); |
2889 | if (!q) { | 2893 | if (!q) { |
2890 | printk(KERN_ERR | 2894 | printk(KERN_ERR |
2891 | "generic_make_request: Trying to access " | 2895 | "generic_make_request: Trying to access " |
2892 | "nonexistent block-device %s (%Lu)\n", | 2896 | "nonexistent block-device %s (%Lu)\n", |
2893 | bdevname(bio->bi_bdev, b), | 2897 | bdevname(bio->bi_bdev, b), |
2894 | (long long) bio->bi_sector); | 2898 | (long long) bio->bi_sector); |
2895 | end_io: | 2899 | end_io: |
2896 | bio_endio(bio, bio->bi_size, -EIO); | 2900 | bio_endio(bio, bio->bi_size, -EIO); |
2897 | break; | 2901 | break; |
2898 | } | 2902 | } |
2899 | 2903 | ||
2900 | if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { | 2904 | if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { |
2901 | printk("bio too big device %s (%u > %u)\n", | 2905 | printk("bio too big device %s (%u > %u)\n", |
2902 | bdevname(bio->bi_bdev, b), | 2906 | bdevname(bio->bi_bdev, b), |
2903 | bio_sectors(bio), | 2907 | bio_sectors(bio), |
2904 | q->max_hw_sectors); | 2908 | q->max_hw_sectors); |
2905 | goto end_io; | 2909 | goto end_io; |
2906 | } | 2910 | } |
2907 | 2911 | ||
2908 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) | 2912 | if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) |
2909 | goto end_io; | 2913 | goto end_io; |
2910 | 2914 | ||
2911 | /* | 2915 | /* |
2912 | * If this device has partitions, remap block n | 2916 | * If this device has partitions, remap block n |
2913 | * of partition p to block n+start(p) of the disk. | 2917 | * of partition p to block n+start(p) of the disk. |
2914 | */ | 2918 | */ |
2915 | blk_partition_remap(bio); | 2919 | blk_partition_remap(bio); |
2916 | 2920 | ||
2917 | ret = q->make_request_fn(q, bio); | 2921 | ret = q->make_request_fn(q, bio); |
2918 | } while (ret); | 2922 | } while (ret); |
2919 | } | 2923 | } |
2920 | 2924 | ||
2921 | EXPORT_SYMBOL(generic_make_request); | 2925 | EXPORT_SYMBOL(generic_make_request); |
2922 | 2926 | ||
2923 | /** | 2927 | /** |
2924 | * submit_bio: submit a bio to the block device layer for I/O | 2928 | * submit_bio: submit a bio to the block device layer for I/O |
2925 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) | 2929 | * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) |
2926 | * @bio: The &struct bio which describes the I/O | 2930 | * @bio: The &struct bio which describes the I/O |
2927 | * | 2931 | * |
2928 | * submit_bio() is very similar in purpose to generic_make_request(), and | 2932 | * submit_bio() is very similar in purpose to generic_make_request(), and |
2929 | * uses that function to do most of the work. Both are fairly rough | 2933 | * uses that function to do most of the work. Both are fairly rough |
2930 | * interfaces, @bio must be presetup and ready for I/O. | 2934 | * interfaces, @bio must be presetup and ready for I/O. |
2931 | * | 2935 | * |
2932 | */ | 2936 | */ |
2933 | void submit_bio(int rw, struct bio *bio) | 2937 | void submit_bio(int rw, struct bio *bio) |
2934 | { | 2938 | { |
2935 | int count = bio_sectors(bio); | 2939 | int count = bio_sectors(bio); |
2936 | 2940 | ||
2937 | BIO_BUG_ON(!bio->bi_size); | 2941 | BIO_BUG_ON(!bio->bi_size); |
2938 | BIO_BUG_ON(!bio->bi_io_vec); | 2942 | BIO_BUG_ON(!bio->bi_io_vec); |
2939 | bio->bi_rw |= rw; | 2943 | bio->bi_rw |= rw; |
2940 | if (rw & WRITE) | 2944 | if (rw & WRITE) |
2941 | mod_page_state(pgpgout, count); | 2945 | mod_page_state(pgpgout, count); |
2942 | else | 2946 | else |
2943 | mod_page_state(pgpgin, count); | 2947 | mod_page_state(pgpgin, count); |
2944 | 2948 | ||
2945 | if (unlikely(block_dump)) { | 2949 | if (unlikely(block_dump)) { |
2946 | char b[BDEVNAME_SIZE]; | 2950 | char b[BDEVNAME_SIZE]; |
2947 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", | 2951 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", |
2948 | current->comm, current->pid, | 2952 | current->comm, current->pid, |
2949 | (rw & WRITE) ? "WRITE" : "READ", | 2953 | (rw & WRITE) ? "WRITE" : "READ", |
2950 | (unsigned long long)bio->bi_sector, | 2954 | (unsigned long long)bio->bi_sector, |
2951 | bdevname(bio->bi_bdev,b)); | 2955 | bdevname(bio->bi_bdev,b)); |
2952 | } | 2956 | } |
2953 | 2957 | ||
2954 | generic_make_request(bio); | 2958 | generic_make_request(bio); |
2955 | } | 2959 | } |
2956 | 2960 | ||
2957 | EXPORT_SYMBOL(submit_bio); | 2961 | EXPORT_SYMBOL(submit_bio); |
2958 | 2962 | ||
2959 | static void blk_recalc_rq_segments(struct request *rq) | 2963 | static void blk_recalc_rq_segments(struct request *rq) |
2960 | { | 2964 | { |
2961 | struct bio *bio, *prevbio = NULL; | 2965 | struct bio *bio, *prevbio = NULL; |
2962 | int nr_phys_segs, nr_hw_segs; | 2966 | int nr_phys_segs, nr_hw_segs; |
2963 | unsigned int phys_size, hw_size; | 2967 | unsigned int phys_size, hw_size; |
2964 | request_queue_t *q = rq->q; | 2968 | request_queue_t *q = rq->q; |
2965 | 2969 | ||
2966 | if (!rq->bio) | 2970 | if (!rq->bio) |
2967 | return; | 2971 | return; |
2968 | 2972 | ||
2969 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; | 2973 | phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; |
2970 | rq_for_each_bio(bio, rq) { | 2974 | rq_for_each_bio(bio, rq) { |
2971 | /* Force bio hw/phys segs to be recalculated. */ | 2975 | /* Force bio hw/phys segs to be recalculated. */ |
2972 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | 2976 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); |
2973 | 2977 | ||
2974 | nr_phys_segs += bio_phys_segments(q, bio); | 2978 | nr_phys_segs += bio_phys_segments(q, bio); |
2975 | nr_hw_segs += bio_hw_segments(q, bio); | 2979 | nr_hw_segs += bio_hw_segments(q, bio); |
2976 | if (prevbio) { | 2980 | if (prevbio) { |
2977 | int pseg = phys_size + prevbio->bi_size + bio->bi_size; | 2981 | int pseg = phys_size + prevbio->bi_size + bio->bi_size; |
2978 | int hseg = hw_size + prevbio->bi_size + bio->bi_size; | 2982 | int hseg = hw_size + prevbio->bi_size + bio->bi_size; |
2979 | 2983 | ||
2980 | if (blk_phys_contig_segment(q, prevbio, bio) && | 2984 | if (blk_phys_contig_segment(q, prevbio, bio) && |
2981 | pseg <= q->max_segment_size) { | 2985 | pseg <= q->max_segment_size) { |
2982 | nr_phys_segs--; | 2986 | nr_phys_segs--; |
2983 | phys_size += prevbio->bi_size + bio->bi_size; | 2987 | phys_size += prevbio->bi_size + bio->bi_size; |
2984 | } else | 2988 | } else |
2985 | phys_size = 0; | 2989 | phys_size = 0; |
2986 | 2990 | ||
2987 | if (blk_hw_contig_segment(q, prevbio, bio) && | 2991 | if (blk_hw_contig_segment(q, prevbio, bio) && |
2988 | hseg <= q->max_segment_size) { | 2992 | hseg <= q->max_segment_size) { |
2989 | nr_hw_segs--; | 2993 | nr_hw_segs--; |
2990 | hw_size += prevbio->bi_size + bio->bi_size; | 2994 | hw_size += prevbio->bi_size + bio->bi_size; |
2991 | } else | 2995 | } else |
2992 | hw_size = 0; | 2996 | hw_size = 0; |
2993 | } | 2997 | } |
2994 | prevbio = bio; | 2998 | prevbio = bio; |
2995 | } | 2999 | } |
2996 | 3000 | ||
2997 | rq->nr_phys_segments = nr_phys_segs; | 3001 | rq->nr_phys_segments = nr_phys_segs; |
2998 | rq->nr_hw_segments = nr_hw_segs; | 3002 | rq->nr_hw_segments = nr_hw_segs; |
2999 | } | 3003 | } |
3000 | 3004 | ||
3001 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) | 3005 | static void blk_recalc_rq_sectors(struct request *rq, int nsect) |
3002 | { | 3006 | { |
3003 | if (blk_fs_request(rq)) { | 3007 | if (blk_fs_request(rq)) { |
3004 | rq->hard_sector += nsect; | 3008 | rq->hard_sector += nsect; |
3005 | rq->hard_nr_sectors -= nsect; | 3009 | rq->hard_nr_sectors -= nsect; |
3006 | 3010 | ||
3007 | /* | 3011 | /* |
3008 | * Move the I/O submission pointers ahead if required. | 3012 | * Move the I/O submission pointers ahead if required. |
3009 | */ | 3013 | */ |
3010 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && | 3014 | if ((rq->nr_sectors >= rq->hard_nr_sectors) && |
3011 | (rq->sector <= rq->hard_sector)) { | 3015 | (rq->sector <= rq->hard_sector)) { |
3012 | rq->sector = rq->hard_sector; | 3016 | rq->sector = rq->hard_sector; |
3013 | rq->nr_sectors = rq->hard_nr_sectors; | 3017 | rq->nr_sectors = rq->hard_nr_sectors; |
3014 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); | 3018 | rq->hard_cur_sectors = bio_cur_sectors(rq->bio); |
3015 | rq->current_nr_sectors = rq->hard_cur_sectors; | 3019 | rq->current_nr_sectors = rq->hard_cur_sectors; |
3016 | rq->buffer = bio_data(rq->bio); | 3020 | rq->buffer = bio_data(rq->bio); |
3017 | } | 3021 | } |
3018 | 3022 | ||
3019 | /* | 3023 | /* |
3020 | * if total number of sectors is less than the first segment | 3024 | * if total number of sectors is less than the first segment |
3021 | * size, something has gone terribly wrong | 3025 | * size, something has gone terribly wrong |
3022 | */ | 3026 | */ |
3023 | if (rq->nr_sectors < rq->current_nr_sectors) { | 3027 | if (rq->nr_sectors < rq->current_nr_sectors) { |
3024 | printk("blk: request botched\n"); | 3028 | printk("blk: request botched\n"); |
3025 | rq->nr_sectors = rq->current_nr_sectors; | 3029 | rq->nr_sectors = rq->current_nr_sectors; |
3026 | } | 3030 | } |
3027 | } | 3031 | } |
3028 | } | 3032 | } |
3029 | 3033 | ||
3030 | static int __end_that_request_first(struct request *req, int uptodate, | 3034 | static int __end_that_request_first(struct request *req, int uptodate, |
3031 | int nr_bytes) | 3035 | int nr_bytes) |
3032 | { | 3036 | { |
3033 | int total_bytes, bio_nbytes, error, next_idx = 0; | 3037 | int total_bytes, bio_nbytes, error, next_idx = 0; |
3034 | struct bio *bio; | 3038 | struct bio *bio; |
3035 | 3039 | ||
3036 | /* | 3040 | /* |
3037 | * extend uptodate bool to allow < 0 value to be direct io error | 3041 | * extend uptodate bool to allow < 0 value to be direct io error |
3038 | */ | 3042 | */ |
3039 | error = 0; | 3043 | error = 0; |
3040 | if (end_io_error(uptodate)) | 3044 | if (end_io_error(uptodate)) |
3041 | error = !uptodate ? -EIO : uptodate; | 3045 | error = !uptodate ? -EIO : uptodate; |
3042 | 3046 | ||
3043 | /* | 3047 | /* |
3044 | * for a REQ_BLOCK_PC request, we want to carry any eventual | 3048 | * for a REQ_BLOCK_PC request, we want to carry any eventual |
3045 | * sense key with us all the way through | 3049 | * sense key with us all the way through |
3046 | */ | 3050 | */ |
3047 | if (!blk_pc_request(req)) | 3051 | if (!blk_pc_request(req)) |
3048 | req->errors = 0; | 3052 | req->errors = 0; |
3049 | 3053 | ||
3050 | if (!uptodate) { | 3054 | if (!uptodate) { |
3051 | if (blk_fs_request(req) && !(req->flags & REQ_QUIET)) | 3055 | if (blk_fs_request(req) && !(req->flags & REQ_QUIET)) |
3052 | printk("end_request: I/O error, dev %s, sector %llu\n", | 3056 | printk("end_request: I/O error, dev %s, sector %llu\n", |
3053 | req->rq_disk ? req->rq_disk->disk_name : "?", | 3057 | req->rq_disk ? req->rq_disk->disk_name : "?", |
3054 | (unsigned long long)req->sector); | 3058 | (unsigned long long)req->sector); |
3055 | } | 3059 | } |
3056 | 3060 | ||
3057 | if (blk_fs_request(req) && req->rq_disk) { | 3061 | if (blk_fs_request(req) && req->rq_disk) { |
3058 | const int rw = rq_data_dir(req); | 3062 | const int rw = rq_data_dir(req); |
3059 | 3063 | ||
3060 | __disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); | 3064 | __disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); |
3061 | } | 3065 | } |
3062 | 3066 | ||
3063 | total_bytes = bio_nbytes = 0; | 3067 | total_bytes = bio_nbytes = 0; |
3064 | while ((bio = req->bio) != NULL) { | 3068 | while ((bio = req->bio) != NULL) { |
3065 | int nbytes; | 3069 | int nbytes; |
3066 | 3070 | ||
3067 | if (nr_bytes >= bio->bi_size) { | 3071 | if (nr_bytes >= bio->bi_size) { |
3068 | req->bio = bio->bi_next; | 3072 | req->bio = bio->bi_next; |
3069 | nbytes = bio->bi_size; | 3073 | nbytes = bio->bi_size; |
3070 | bio_endio(bio, nbytes, error); | 3074 | bio_endio(bio, nbytes, error); |
3071 | next_idx = 0; | 3075 | next_idx = 0; |
3072 | bio_nbytes = 0; | 3076 | bio_nbytes = 0; |
3073 | } else { | 3077 | } else { |
3074 | int idx = bio->bi_idx + next_idx; | 3078 | int idx = bio->bi_idx + next_idx; |
3075 | 3079 | ||
3076 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { | 3080 | if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { |
3077 | blk_dump_rq_flags(req, "__end_that"); | 3081 | blk_dump_rq_flags(req, "__end_that"); |
3078 | printk("%s: bio idx %d >= vcnt %d\n", | 3082 | printk("%s: bio idx %d >= vcnt %d\n", |
3079 | __FUNCTION__, | 3083 | __FUNCTION__, |
3080 | bio->bi_idx, bio->bi_vcnt); | 3084 | bio->bi_idx, bio->bi_vcnt); |
3081 | break; | 3085 | break; |
3082 | } | 3086 | } |
3083 | 3087 | ||
3084 | nbytes = bio_iovec_idx(bio, idx)->bv_len; | 3088 | nbytes = bio_iovec_idx(bio, idx)->bv_len; |
3085 | BIO_BUG_ON(nbytes > bio->bi_size); | 3089 | BIO_BUG_ON(nbytes > bio->bi_size); |
3086 | 3090 | ||
3087 | /* | 3091 | /* |
3088 | * not a complete bvec done | 3092 | * not a complete bvec done |
3089 | */ | 3093 | */ |
3090 | if (unlikely(nbytes > nr_bytes)) { | 3094 | if (unlikely(nbytes > nr_bytes)) { |
3091 | bio_nbytes += nr_bytes; | 3095 | bio_nbytes += nr_bytes; |
3092 | total_bytes += nr_bytes; | 3096 | total_bytes += nr_bytes; |
3093 | break; | 3097 | break; |
3094 | } | 3098 | } |
3095 | 3099 | ||
3096 | /* | 3100 | /* |
3097 | * advance to the next vector | 3101 | * advance to the next vector |
3098 | */ | 3102 | */ |
3099 | next_idx++; | 3103 | next_idx++; |
3100 | bio_nbytes += nbytes; | 3104 | bio_nbytes += nbytes; |
3101 | } | 3105 | } |
3102 | 3106 | ||
3103 | total_bytes += nbytes; | 3107 | total_bytes += nbytes; |
3104 | nr_bytes -= nbytes; | 3108 | nr_bytes -= nbytes; |
3105 | 3109 | ||
3106 | if ((bio = req->bio)) { | 3110 | if ((bio = req->bio)) { |
3107 | /* | 3111 | /* |
3108 | * end more in this run, or just return 'not-done' | 3112 | * end more in this run, or just return 'not-done' |
3109 | */ | 3113 | */ |
3110 | if (unlikely(nr_bytes <= 0)) | 3114 | if (unlikely(nr_bytes <= 0)) |
3111 | break; | 3115 | break; |
3112 | } | 3116 | } |
3113 | } | 3117 | } |
3114 | 3118 | ||
3115 | /* | 3119 | /* |
3116 | * completely done | 3120 | * completely done |
3117 | */ | 3121 | */ |
3118 | if (!req->bio) | 3122 | if (!req->bio) |
3119 | return 0; | 3123 | return 0; |
3120 | 3124 | ||
3121 | /* | 3125 | /* |
3122 | * if the request wasn't completed, update state | 3126 | * if the request wasn't completed, update state |
3123 | */ | 3127 | */ |
3124 | if (bio_nbytes) { | 3128 | if (bio_nbytes) { |
3125 | bio_endio(bio, bio_nbytes, error); | 3129 | bio_endio(bio, bio_nbytes, error); |
3126 | bio->bi_idx += next_idx; | 3130 | bio->bi_idx += next_idx; |
3127 | bio_iovec(bio)->bv_offset += nr_bytes; | 3131 | bio_iovec(bio)->bv_offset += nr_bytes; |
3128 | bio_iovec(bio)->bv_len -= nr_bytes; | 3132 | bio_iovec(bio)->bv_len -= nr_bytes; |
3129 | } | 3133 | } |
3130 | 3134 | ||
3131 | blk_recalc_rq_sectors(req, total_bytes >> 9); | 3135 | blk_recalc_rq_sectors(req, total_bytes >> 9); |
3132 | blk_recalc_rq_segments(req); | 3136 | blk_recalc_rq_segments(req); |
3133 | return 1; | 3137 | return 1; |
3134 | } | 3138 | } |
3135 | 3139 | ||
3136 | /** | 3140 | /** |
3137 | * end_that_request_first - end I/O on a request | 3141 | * end_that_request_first - end I/O on a request |
3138 | * @req: the request being processed | 3142 | * @req: the request being processed |
3139 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | 3143 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error |
3140 | * @nr_sectors: number of sectors to end I/O on | 3144 | * @nr_sectors: number of sectors to end I/O on |
3141 | * | 3145 | * |
3142 | * Description: | 3146 | * Description: |
3143 | * Ends I/O on a number of sectors attached to @req, and sets it up | 3147 | * Ends I/O on a number of sectors attached to @req, and sets it up |
3144 | * for the next range of segments (if any) in the cluster. | 3148 | * for the next range of segments (if any) in the cluster. |
3145 | * | 3149 | * |
3146 | * Return: | 3150 | * Return: |
3147 | * 0 - we are done with this request, call end_that_request_last() | 3151 | * 0 - we are done with this request, call end_that_request_last() |
3148 | * 1 - still buffers pending for this request | 3152 | * 1 - still buffers pending for this request |
3149 | **/ | 3153 | **/ |
3150 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) | 3154 | int end_that_request_first(struct request *req, int uptodate, int nr_sectors) |
3151 | { | 3155 | { |
3152 | return __end_that_request_first(req, uptodate, nr_sectors << 9); | 3156 | return __end_that_request_first(req, uptodate, nr_sectors << 9); |
3153 | } | 3157 | } |
3154 | 3158 | ||
3155 | EXPORT_SYMBOL(end_that_request_first); | 3159 | EXPORT_SYMBOL(end_that_request_first); |
3156 | 3160 | ||
3157 | /** | 3161 | /** |
3158 | * end_that_request_chunk - end I/O on a request | 3162 | * end_that_request_chunk - end I/O on a request |
3159 | * @req: the request being processed | 3163 | * @req: the request being processed |
3160 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error | 3164 | * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error |
3161 | * @nr_bytes: number of bytes to complete | 3165 | * @nr_bytes: number of bytes to complete |
3162 | * | 3166 | * |
3163 | * Description: | 3167 | * Description: |
3164 | * Ends I/O on a number of bytes attached to @req, and sets it up | 3168 | * Ends I/O on a number of bytes attached to @req, and sets it up |
3165 | * for the next range of segments (if any). Like end_that_request_first(), | 3169 | * for the next range of segments (if any). Like end_that_request_first(), |
3166 | * but deals with bytes instead of sectors. | 3170 | * but deals with bytes instead of sectors. |
3167 | * | 3171 | * |
3168 | * Return: | 3172 | * Return: |
3169 | * 0 - we are done with this request, call end_that_request_last() | 3173 | * 0 - we are done with this request, call end_that_request_last() |
3170 | * 1 - still buffers pending for this request | 3174 | * 1 - still buffers pending for this request |
3171 | **/ | 3175 | **/ |
3172 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) | 3176 | int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) |
3173 | { | 3177 | { |
3174 | return __end_that_request_first(req, uptodate, nr_bytes); | 3178 | return __end_that_request_first(req, uptodate, nr_bytes); |
3175 | } | 3179 | } |
3176 | 3180 | ||
3177 | EXPORT_SYMBOL(end_that_request_chunk); | 3181 | EXPORT_SYMBOL(end_that_request_chunk); |
3178 | 3182 | ||
3179 | /* | 3183 | /* |
3180 | * queue lock must be held | 3184 | * queue lock must be held |
3181 | */ | 3185 | */ |
3182 | void end_that_request_last(struct request *req) | 3186 | void end_that_request_last(struct request *req) |
3183 | { | 3187 | { |
3184 | struct gendisk *disk = req->rq_disk; | 3188 | struct gendisk *disk = req->rq_disk; |
3185 | 3189 | ||
3186 | if (unlikely(laptop_mode) && blk_fs_request(req)) | 3190 | if (unlikely(laptop_mode) && blk_fs_request(req)) |
3187 | laptop_io_completion(); | 3191 | laptop_io_completion(); |
3188 | 3192 | ||
3189 | if (disk && blk_fs_request(req)) { | 3193 | if (disk && blk_fs_request(req)) { |
3190 | unsigned long duration = jiffies - req->start_time; | 3194 | unsigned long duration = jiffies - req->start_time; |
3191 | const int rw = rq_data_dir(req); | 3195 | const int rw = rq_data_dir(req); |
3192 | 3196 | ||
3193 | __disk_stat_inc(disk, ios[rw]); | 3197 | __disk_stat_inc(disk, ios[rw]); |
3194 | __disk_stat_add(disk, ticks[rw], duration); | 3198 | __disk_stat_add(disk, ticks[rw], duration); |
3195 | disk_round_stats(disk); | 3199 | disk_round_stats(disk); |
3196 | disk->in_flight--; | 3200 | disk->in_flight--; |
3197 | } | 3201 | } |
3198 | if (req->end_io) | 3202 | if (req->end_io) |
3199 | req->end_io(req); | 3203 | req->end_io(req); |
3200 | else | 3204 | else |
3201 | __blk_put_request(req->q, req); | 3205 | __blk_put_request(req->q, req); |
3202 | } | 3206 | } |
3203 | 3207 | ||
3204 | EXPORT_SYMBOL(end_that_request_last); | 3208 | EXPORT_SYMBOL(end_that_request_last); |
3205 | 3209 | ||
3206 | void end_request(struct request *req, int uptodate) | 3210 | void end_request(struct request *req, int uptodate) |
3207 | { | 3211 | { |
3208 | if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { | 3212 | if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { |
3209 | add_disk_randomness(req->rq_disk); | 3213 | add_disk_randomness(req->rq_disk); |
3210 | blkdev_dequeue_request(req); | 3214 | blkdev_dequeue_request(req); |
3211 | end_that_request_last(req); | 3215 | end_that_request_last(req); |
3212 | } | 3216 | } |
3213 | } | 3217 | } |
3214 | 3218 | ||
3215 | EXPORT_SYMBOL(end_request); | 3219 | EXPORT_SYMBOL(end_request); |
3216 | 3220 | ||
3217 | void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio) | 3221 | void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio) |
3218 | { | 3222 | { |
3219 | /* first three bits are identical in rq->flags and bio->bi_rw */ | 3223 | /* first three bits are identical in rq->flags and bio->bi_rw */ |
3220 | rq->flags |= (bio->bi_rw & 7); | 3224 | rq->flags |= (bio->bi_rw & 7); |
3221 | 3225 | ||
3222 | rq->nr_phys_segments = bio_phys_segments(q, bio); | 3226 | rq->nr_phys_segments = bio_phys_segments(q, bio); |
3223 | rq->nr_hw_segments = bio_hw_segments(q, bio); | 3227 | rq->nr_hw_segments = bio_hw_segments(q, bio); |
3224 | rq->current_nr_sectors = bio_cur_sectors(bio); | 3228 | rq->current_nr_sectors = bio_cur_sectors(bio); |
3225 | rq->hard_cur_sectors = rq->current_nr_sectors; | 3229 | rq->hard_cur_sectors = rq->current_nr_sectors; |
3226 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); | 3230 | rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); |
3227 | rq->buffer = bio_data(bio); | 3231 | rq->buffer = bio_data(bio); |
3228 | 3232 | ||
3229 | rq->bio = rq->biotail = bio; | 3233 | rq->bio = rq->biotail = bio; |
3230 | } | 3234 | } |
3231 | 3235 | ||
3232 | EXPORT_SYMBOL(blk_rq_bio_prep); | 3236 | EXPORT_SYMBOL(blk_rq_bio_prep); |
3233 | 3237 | ||
3234 | int kblockd_schedule_work(struct work_struct *work) | 3238 | int kblockd_schedule_work(struct work_struct *work) |
3235 | { | 3239 | { |
3236 | return queue_work(kblockd_workqueue, work); | 3240 | return queue_work(kblockd_workqueue, work); |
3237 | } | 3241 | } |
3238 | 3242 | ||
3239 | EXPORT_SYMBOL(kblockd_schedule_work); | 3243 | EXPORT_SYMBOL(kblockd_schedule_work); |
3240 | 3244 | ||
3241 | void kblockd_flush(void) | 3245 | void kblockd_flush(void) |
3242 | { | 3246 | { |
3243 | flush_workqueue(kblockd_workqueue); | 3247 | flush_workqueue(kblockd_workqueue); |
3244 | } | 3248 | } |
3245 | EXPORT_SYMBOL(kblockd_flush); | 3249 | EXPORT_SYMBOL(kblockd_flush); |
3246 | 3250 | ||
3247 | int __init blk_dev_init(void) | 3251 | int __init blk_dev_init(void) |
3248 | { | 3252 | { |
3249 | kblockd_workqueue = create_workqueue("kblockd"); | 3253 | kblockd_workqueue = create_workqueue("kblockd"); |
3250 | if (!kblockd_workqueue) | 3254 | if (!kblockd_workqueue) |
3251 | panic("Failed to create kblockd\n"); | 3255 | panic("Failed to create kblockd\n"); |
3252 | 3256 | ||
3253 | request_cachep = kmem_cache_create("blkdev_requests", | 3257 | request_cachep = kmem_cache_create("blkdev_requests", |
3254 | sizeof(struct request), 0, SLAB_PANIC, NULL, NULL); | 3258 | sizeof(struct request), 0, SLAB_PANIC, NULL, NULL); |
3255 | 3259 | ||
3256 | requestq_cachep = kmem_cache_create("blkdev_queue", | 3260 | requestq_cachep = kmem_cache_create("blkdev_queue", |
3257 | sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL); | 3261 | sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL); |
3258 | 3262 | ||
3259 | iocontext_cachep = kmem_cache_create("blkdev_ioc", | 3263 | iocontext_cachep = kmem_cache_create("blkdev_ioc", |
3260 | sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL); | 3264 | sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL); |
3261 | 3265 | ||
3262 | blk_max_low_pfn = max_low_pfn; | 3266 | blk_max_low_pfn = max_low_pfn; |
3263 | blk_max_pfn = max_pfn; | 3267 | blk_max_pfn = max_pfn; |
3264 | 3268 | ||
3265 | return 0; | 3269 | return 0; |
3266 | } | 3270 | } |
3267 | 3271 | ||
3268 | /* | 3272 | /* |
3269 | * IO Context helper functions | 3273 | * IO Context helper functions |
3270 | */ | 3274 | */ |
3271 | void put_io_context(struct io_context *ioc) | 3275 | void put_io_context(struct io_context *ioc) |
3272 | { | 3276 | { |
3273 | if (ioc == NULL) | 3277 | if (ioc == NULL) |
3274 | return; | 3278 | return; |
3275 | 3279 | ||
3276 | BUG_ON(atomic_read(&ioc->refcount) == 0); | 3280 | BUG_ON(atomic_read(&ioc->refcount) == 0); |
3277 | 3281 | ||
3278 | if (atomic_dec_and_test(&ioc->refcount)) { | 3282 | if (atomic_dec_and_test(&ioc->refcount)) { |
3279 | if (ioc->aic && ioc->aic->dtor) | 3283 | if (ioc->aic && ioc->aic->dtor) |
3280 | ioc->aic->dtor(ioc->aic); | 3284 | ioc->aic->dtor(ioc->aic); |
3281 | if (ioc->cic && ioc->cic->dtor) | 3285 | if (ioc->cic && ioc->cic->dtor) |
3282 | ioc->cic->dtor(ioc->cic); | 3286 | ioc->cic->dtor(ioc->cic); |
3283 | 3287 | ||
3284 | kmem_cache_free(iocontext_cachep, ioc); | 3288 | kmem_cache_free(iocontext_cachep, ioc); |
3285 | } | 3289 | } |
3286 | } | 3290 | } |
3287 | EXPORT_SYMBOL(put_io_context); | 3291 | EXPORT_SYMBOL(put_io_context); |
3288 | 3292 | ||
3289 | /* Called by the exitting task */ | 3293 | /* Called by the exitting task */ |
3290 | void exit_io_context(void) | 3294 | void exit_io_context(void) |
3291 | { | 3295 | { |
3292 | unsigned long flags; | 3296 | unsigned long flags; |
3293 | struct io_context *ioc; | 3297 | struct io_context *ioc; |
3294 | 3298 | ||
3295 | local_irq_save(flags); | 3299 | local_irq_save(flags); |
3296 | task_lock(current); | 3300 | task_lock(current); |
3297 | ioc = current->io_context; | 3301 | ioc = current->io_context; |
3298 | current->io_context = NULL; | 3302 | current->io_context = NULL; |
3299 | ioc->task = NULL; | 3303 | ioc->task = NULL; |
3300 | task_unlock(current); | 3304 | task_unlock(current); |
3301 | local_irq_restore(flags); | 3305 | local_irq_restore(flags); |
3302 | 3306 | ||
3303 | if (ioc->aic && ioc->aic->exit) | 3307 | if (ioc->aic && ioc->aic->exit) |
3304 | ioc->aic->exit(ioc->aic); | 3308 | ioc->aic->exit(ioc->aic); |
3305 | if (ioc->cic && ioc->cic->exit) | 3309 | if (ioc->cic && ioc->cic->exit) |
3306 | ioc->cic->exit(ioc->cic); | 3310 | ioc->cic->exit(ioc->cic); |
3307 | 3311 | ||
3308 | put_io_context(ioc); | 3312 | put_io_context(ioc); |
3309 | } | 3313 | } |
3310 | 3314 | ||
3311 | /* | 3315 | /* |
3312 | * If the current task has no IO context then create one and initialise it. | 3316 | * If the current task has no IO context then create one and initialise it. |
3313 | * Otherwise, return its existing IO context. | 3317 | * Otherwise, return its existing IO context. |
3314 | * | 3318 | * |
3315 | * This returned IO context doesn't have a specifically elevated refcount, | 3319 | * This returned IO context doesn't have a specifically elevated refcount, |
3316 | * but since the current task itself holds a reference, the context can be | 3320 | * but since the current task itself holds a reference, the context can be |
3317 | * used in general code, so long as it stays within `current` context. | 3321 | * used in general code, so long as it stays within `current` context. |
3318 | */ | 3322 | */ |
3319 | struct io_context *current_io_context(gfp_t gfp_flags) | 3323 | struct io_context *current_io_context(gfp_t gfp_flags) |
3320 | { | 3324 | { |
3321 | struct task_struct *tsk = current; | 3325 | struct task_struct *tsk = current; |
3322 | struct io_context *ret; | 3326 | struct io_context *ret; |
3323 | 3327 | ||
3324 | ret = tsk->io_context; | 3328 | ret = tsk->io_context; |
3325 | if (likely(ret)) | 3329 | if (likely(ret)) |
3326 | return ret; | 3330 | return ret; |
3327 | 3331 | ||
3328 | ret = kmem_cache_alloc(iocontext_cachep, gfp_flags); | 3332 | ret = kmem_cache_alloc(iocontext_cachep, gfp_flags); |
3329 | if (ret) { | 3333 | if (ret) { |
3330 | atomic_set(&ret->refcount, 1); | 3334 | atomic_set(&ret->refcount, 1); |
3331 | ret->task = current; | 3335 | ret->task = current; |
3332 | ret->set_ioprio = NULL; | 3336 | ret->set_ioprio = NULL; |
3333 | ret->last_waited = jiffies; /* doesn't matter... */ | 3337 | ret->last_waited = jiffies; /* doesn't matter... */ |
3334 | ret->nr_batch_requests = 0; /* because this is 0 */ | 3338 | ret->nr_batch_requests = 0; /* because this is 0 */ |
3335 | ret->aic = NULL; | 3339 | ret->aic = NULL; |
3336 | ret->cic = NULL; | 3340 | ret->cic = NULL; |
3337 | tsk->io_context = ret; | 3341 | tsk->io_context = ret; |
3338 | } | 3342 | } |
3339 | 3343 | ||
3340 | return ret; | 3344 | return ret; |
3341 | } | 3345 | } |
3342 | EXPORT_SYMBOL(current_io_context); | 3346 | EXPORT_SYMBOL(current_io_context); |
3343 | 3347 | ||
3344 | /* | 3348 | /* |
3345 | * If the current task has no IO context then create one and initialise it. | 3349 | * If the current task has no IO context then create one and initialise it. |
3346 | * If it does have a context, take a ref on it. | 3350 | * If it does have a context, take a ref on it. |
3347 | * | 3351 | * |
3348 | * This is always called in the context of the task which submitted the I/O. | 3352 | * This is always called in the context of the task which submitted the I/O. |
3349 | */ | 3353 | */ |
3350 | struct io_context *get_io_context(gfp_t gfp_flags) | 3354 | struct io_context *get_io_context(gfp_t gfp_flags) |
3351 | { | 3355 | { |
3352 | struct io_context *ret; | 3356 | struct io_context *ret; |
3353 | ret = current_io_context(gfp_flags); | 3357 | ret = current_io_context(gfp_flags); |
3354 | if (likely(ret)) | 3358 | if (likely(ret)) |
3355 | atomic_inc(&ret->refcount); | 3359 | atomic_inc(&ret->refcount); |
3356 | return ret; | 3360 | return ret; |
3357 | } | 3361 | } |
3358 | EXPORT_SYMBOL(get_io_context); | 3362 | EXPORT_SYMBOL(get_io_context); |
3359 | 3363 | ||
3360 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) | 3364 | void copy_io_context(struct io_context **pdst, struct io_context **psrc) |
3361 | { | 3365 | { |
3362 | struct io_context *src = *psrc; | 3366 | struct io_context *src = *psrc; |
3363 | struct io_context *dst = *pdst; | 3367 | struct io_context *dst = *pdst; |
3364 | 3368 | ||
3365 | if (src) { | 3369 | if (src) { |
3366 | BUG_ON(atomic_read(&src->refcount) == 0); | 3370 | BUG_ON(atomic_read(&src->refcount) == 0); |
3367 | atomic_inc(&src->refcount); | 3371 | atomic_inc(&src->refcount); |
3368 | put_io_context(dst); | 3372 | put_io_context(dst); |
3369 | *pdst = src; | 3373 | *pdst = src; |
3370 | } | 3374 | } |
3371 | } | 3375 | } |
3372 | EXPORT_SYMBOL(copy_io_context); | 3376 | EXPORT_SYMBOL(copy_io_context); |
3373 | 3377 | ||
3374 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) | 3378 | void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) |
3375 | { | 3379 | { |
3376 | struct io_context *temp; | 3380 | struct io_context *temp; |
3377 | temp = *ioc1; | 3381 | temp = *ioc1; |
3378 | *ioc1 = *ioc2; | 3382 | *ioc1 = *ioc2; |
3379 | *ioc2 = temp; | 3383 | *ioc2 = temp; |
3380 | } | 3384 | } |
3381 | EXPORT_SYMBOL(swap_io_context); | 3385 | EXPORT_SYMBOL(swap_io_context); |
3382 | 3386 | ||
3383 | /* | 3387 | /* |
3384 | * sysfs parts below | 3388 | * sysfs parts below |
3385 | */ | 3389 | */ |
3386 | struct queue_sysfs_entry { | 3390 | struct queue_sysfs_entry { |
3387 | struct attribute attr; | 3391 | struct attribute attr; |
3388 | ssize_t (*show)(struct request_queue *, char *); | 3392 | ssize_t (*show)(struct request_queue *, char *); |
3389 | ssize_t (*store)(struct request_queue *, const char *, size_t); | 3393 | ssize_t (*store)(struct request_queue *, const char *, size_t); |
3390 | }; | 3394 | }; |
3391 | 3395 | ||
3392 | static ssize_t | 3396 | static ssize_t |
3393 | queue_var_show(unsigned int var, char *page) | 3397 | queue_var_show(unsigned int var, char *page) |
3394 | { | 3398 | { |
3395 | return sprintf(page, "%d\n", var); | 3399 | return sprintf(page, "%d\n", var); |
3396 | } | 3400 | } |
3397 | 3401 | ||
3398 | static ssize_t | 3402 | static ssize_t |
3399 | queue_var_store(unsigned long *var, const char *page, size_t count) | 3403 | queue_var_store(unsigned long *var, const char *page, size_t count) |
3400 | { | 3404 | { |
3401 | char *p = (char *) page; | 3405 | char *p = (char *) page; |
3402 | 3406 | ||
3403 | *var = simple_strtoul(p, &p, 10); | 3407 | *var = simple_strtoul(p, &p, 10); |
3404 | return count; | 3408 | return count; |
3405 | } | 3409 | } |
3406 | 3410 | ||
3407 | static ssize_t queue_requests_show(struct request_queue *q, char *page) | 3411 | static ssize_t queue_requests_show(struct request_queue *q, char *page) |
3408 | { | 3412 | { |
3409 | return queue_var_show(q->nr_requests, (page)); | 3413 | return queue_var_show(q->nr_requests, (page)); |
3410 | } | 3414 | } |
3411 | 3415 | ||
3412 | static ssize_t | 3416 | static ssize_t |
3413 | queue_requests_store(struct request_queue *q, const char *page, size_t count) | 3417 | queue_requests_store(struct request_queue *q, const char *page, size_t count) |
3414 | { | 3418 | { |
3415 | struct request_list *rl = &q->rq; | 3419 | struct request_list *rl = &q->rq; |
3416 | 3420 | ||
3417 | int ret = queue_var_store(&q->nr_requests, page, count); | 3421 | int ret = queue_var_store(&q->nr_requests, page, count); |
3418 | if (q->nr_requests < BLKDEV_MIN_RQ) | 3422 | if (q->nr_requests < BLKDEV_MIN_RQ) |
3419 | q->nr_requests = BLKDEV_MIN_RQ; | 3423 | q->nr_requests = BLKDEV_MIN_RQ; |
3420 | blk_queue_congestion_threshold(q); | 3424 | blk_queue_congestion_threshold(q); |
3421 | 3425 | ||
3422 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) | 3426 | if (rl->count[READ] >= queue_congestion_on_threshold(q)) |
3423 | set_queue_congested(q, READ); | 3427 | set_queue_congested(q, READ); |
3424 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) | 3428 | else if (rl->count[READ] < queue_congestion_off_threshold(q)) |
3425 | clear_queue_congested(q, READ); | 3429 | clear_queue_congested(q, READ); |
3426 | 3430 | ||
3427 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) | 3431 | if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) |
3428 | set_queue_congested(q, WRITE); | 3432 | set_queue_congested(q, WRITE); |
3429 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) | 3433 | else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) |
3430 | clear_queue_congested(q, WRITE); | 3434 | clear_queue_congested(q, WRITE); |
3431 | 3435 | ||
3432 | if (rl->count[READ] >= q->nr_requests) { | 3436 | if (rl->count[READ] >= q->nr_requests) { |
3433 | blk_set_queue_full(q, READ); | 3437 | blk_set_queue_full(q, READ); |
3434 | } else if (rl->count[READ]+1 <= q->nr_requests) { | 3438 | } else if (rl->count[READ]+1 <= q->nr_requests) { |
3435 | blk_clear_queue_full(q, READ); | 3439 | blk_clear_queue_full(q, READ); |
3436 | wake_up(&rl->wait[READ]); | 3440 | wake_up(&rl->wait[READ]); |
3437 | } | 3441 | } |
3438 | 3442 | ||
3439 | if (rl->count[WRITE] >= q->nr_requests) { | 3443 | if (rl->count[WRITE] >= q->nr_requests) { |
3440 | blk_set_queue_full(q, WRITE); | 3444 | blk_set_queue_full(q, WRITE); |
3441 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { | 3445 | } else if (rl->count[WRITE]+1 <= q->nr_requests) { |
3442 | blk_clear_queue_full(q, WRITE); | 3446 | blk_clear_queue_full(q, WRITE); |
3443 | wake_up(&rl->wait[WRITE]); | 3447 | wake_up(&rl->wait[WRITE]); |
3444 | } | 3448 | } |
3445 | return ret; | 3449 | return ret; |
3446 | } | 3450 | } |
3447 | 3451 | ||
3448 | static ssize_t queue_ra_show(struct request_queue *q, char *page) | 3452 | static ssize_t queue_ra_show(struct request_queue *q, char *page) |
3449 | { | 3453 | { |
3450 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | 3454 | int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); |
3451 | 3455 | ||
3452 | return queue_var_show(ra_kb, (page)); | 3456 | return queue_var_show(ra_kb, (page)); |
3453 | } | 3457 | } |
3454 | 3458 | ||
3455 | static ssize_t | 3459 | static ssize_t |
3456 | queue_ra_store(struct request_queue *q, const char *page, size_t count) | 3460 | queue_ra_store(struct request_queue *q, const char *page, size_t count) |
3457 | { | 3461 | { |
3458 | unsigned long ra_kb; | 3462 | unsigned long ra_kb; |
3459 | ssize_t ret = queue_var_store(&ra_kb, page, count); | 3463 | ssize_t ret = queue_var_store(&ra_kb, page, count); |
3460 | 3464 | ||
3461 | spin_lock_irq(q->queue_lock); | 3465 | spin_lock_irq(q->queue_lock); |
3462 | if (ra_kb > (q->max_sectors >> 1)) | 3466 | if (ra_kb > (q->max_sectors >> 1)) |
3463 | ra_kb = (q->max_sectors >> 1); | 3467 | ra_kb = (q->max_sectors >> 1); |
3464 | 3468 | ||
3465 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); | 3469 | q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); |
3466 | spin_unlock_irq(q->queue_lock); | 3470 | spin_unlock_irq(q->queue_lock); |
3467 | 3471 | ||
3468 | return ret; | 3472 | return ret; |
3469 | } | 3473 | } |
3470 | 3474 | ||
3471 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) | 3475 | static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) |
3472 | { | 3476 | { |
3473 | int max_sectors_kb = q->max_sectors >> 1; | 3477 | int max_sectors_kb = q->max_sectors >> 1; |
3474 | 3478 | ||
3475 | return queue_var_show(max_sectors_kb, (page)); | 3479 | return queue_var_show(max_sectors_kb, (page)); |
3476 | } | 3480 | } |
3477 | 3481 | ||
3478 | static ssize_t | 3482 | static ssize_t |
3479 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) | 3483 | queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) |
3480 | { | 3484 | { |
3481 | unsigned long max_sectors_kb, | 3485 | unsigned long max_sectors_kb, |
3482 | max_hw_sectors_kb = q->max_hw_sectors >> 1, | 3486 | max_hw_sectors_kb = q->max_hw_sectors >> 1, |
3483 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); | 3487 | page_kb = 1 << (PAGE_CACHE_SHIFT - 10); |
3484 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); | 3488 | ssize_t ret = queue_var_store(&max_sectors_kb, page, count); |
3485 | int ra_kb; | 3489 | int ra_kb; |
3486 | 3490 | ||
3487 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) | 3491 | if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) |
3488 | return -EINVAL; | 3492 | return -EINVAL; |
3489 | /* | 3493 | /* |
3490 | * Take the queue lock to update the readahead and max_sectors | 3494 | * Take the queue lock to update the readahead and max_sectors |
3491 | * values synchronously: | 3495 | * values synchronously: |
3492 | */ | 3496 | */ |
3493 | spin_lock_irq(q->queue_lock); | 3497 | spin_lock_irq(q->queue_lock); |
3494 | /* | 3498 | /* |
3495 | * Trim readahead window as well, if necessary: | 3499 | * Trim readahead window as well, if necessary: |
3496 | */ | 3500 | */ |
3497 | ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); | 3501 | ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); |
3498 | if (ra_kb > max_sectors_kb) | 3502 | if (ra_kb > max_sectors_kb) |
3499 | q->backing_dev_info.ra_pages = | 3503 | q->backing_dev_info.ra_pages = |
3500 | max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); | 3504 | max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); |
3501 | 3505 | ||
3502 | q->max_sectors = max_sectors_kb << 1; | 3506 | q->max_sectors = max_sectors_kb << 1; |
3503 | spin_unlock_irq(q->queue_lock); | 3507 | spin_unlock_irq(q->queue_lock); |
3504 | 3508 | ||
3505 | return ret; | 3509 | return ret; |
3506 | } | 3510 | } |
3507 | 3511 | ||
3508 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) | 3512 | static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) |
3509 | { | 3513 | { |
3510 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; | 3514 | int max_hw_sectors_kb = q->max_hw_sectors >> 1; |
3511 | 3515 | ||
3512 | return queue_var_show(max_hw_sectors_kb, (page)); | 3516 | return queue_var_show(max_hw_sectors_kb, (page)); |
3513 | } | 3517 | } |
3514 | 3518 | ||
3515 | 3519 | ||
3516 | static struct queue_sysfs_entry queue_requests_entry = { | 3520 | static struct queue_sysfs_entry queue_requests_entry = { |
3517 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, | 3521 | .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, |
3518 | .show = queue_requests_show, | 3522 | .show = queue_requests_show, |
3519 | .store = queue_requests_store, | 3523 | .store = queue_requests_store, |
3520 | }; | 3524 | }; |
3521 | 3525 | ||
3522 | static struct queue_sysfs_entry queue_ra_entry = { | 3526 | static struct queue_sysfs_entry queue_ra_entry = { |
3523 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, | 3527 | .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, |
3524 | .show = queue_ra_show, | 3528 | .show = queue_ra_show, |
3525 | .store = queue_ra_store, | 3529 | .store = queue_ra_store, |
3526 | }; | 3530 | }; |
3527 | 3531 | ||
3528 | static struct queue_sysfs_entry queue_max_sectors_entry = { | 3532 | static struct queue_sysfs_entry queue_max_sectors_entry = { |
3529 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, | 3533 | .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, |
3530 | .show = queue_max_sectors_show, | 3534 | .show = queue_max_sectors_show, |
3531 | .store = queue_max_sectors_store, | 3535 | .store = queue_max_sectors_store, |
3532 | }; | 3536 | }; |
3533 | 3537 | ||
3534 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { | 3538 | static struct queue_sysfs_entry queue_max_hw_sectors_entry = { |
3535 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, | 3539 | .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, |
3536 | .show = queue_max_hw_sectors_show, | 3540 | .show = queue_max_hw_sectors_show, |
3537 | }; | 3541 | }; |
3538 | 3542 | ||
3539 | static struct queue_sysfs_entry queue_iosched_entry = { | 3543 | static struct queue_sysfs_entry queue_iosched_entry = { |
3540 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, | 3544 | .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, |
3541 | .show = elv_iosched_show, | 3545 | .show = elv_iosched_show, |
3542 | .store = elv_iosched_store, | 3546 | .store = elv_iosched_store, |
3543 | }; | 3547 | }; |
3544 | 3548 | ||
3545 | static struct attribute *default_attrs[] = { | 3549 | static struct attribute *default_attrs[] = { |
3546 | &queue_requests_entry.attr, | 3550 | &queue_requests_entry.attr, |
3547 | &queue_ra_entry.attr, | 3551 | &queue_ra_entry.attr, |
3548 | &queue_max_hw_sectors_entry.attr, | 3552 | &queue_max_hw_sectors_entry.attr, |
3549 | &queue_max_sectors_entry.attr, | 3553 | &queue_max_sectors_entry.attr, |
3550 | &queue_iosched_entry.attr, | 3554 | &queue_iosched_entry.attr, |
3551 | NULL, | 3555 | NULL, |
3552 | }; | 3556 | }; |
3553 | 3557 | ||
3554 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) | 3558 | #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) |
3555 | 3559 | ||
3556 | static ssize_t | 3560 | static ssize_t |
3557 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | 3561 | queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) |
3558 | { | 3562 | { |
3559 | struct queue_sysfs_entry *entry = to_queue(attr); | 3563 | struct queue_sysfs_entry *entry = to_queue(attr); |
3560 | struct request_queue *q; | 3564 | struct request_queue *q; |
3561 | 3565 | ||
3562 | q = container_of(kobj, struct request_queue, kobj); | 3566 | q = container_of(kobj, struct request_queue, kobj); |
3563 | if (!entry->show) | 3567 | if (!entry->show) |
3564 | return -EIO; | 3568 | return -EIO; |
3565 | 3569 | ||
3566 | return entry->show(q, page); | 3570 | return entry->show(q, page); |
3567 | } | 3571 | } |
3568 | 3572 | ||
3569 | static ssize_t | 3573 | static ssize_t |
3570 | queue_attr_store(struct kobject *kobj, struct attribute *attr, | 3574 | queue_attr_store(struct kobject *kobj, struct attribute *attr, |
3571 | const char *page, size_t length) | 3575 | const char *page, size_t length) |
3572 | { | 3576 | { |
3573 | struct queue_sysfs_entry *entry = to_queue(attr); | 3577 | struct queue_sysfs_entry *entry = to_queue(attr); |
3574 | struct request_queue *q; | 3578 | struct request_queue *q; |
3575 | 3579 | ||
3576 | q = container_of(kobj, struct request_queue, kobj); | 3580 | q = container_of(kobj, struct request_queue, kobj); |
3577 | if (!entry->store) | 3581 | if (!entry->store) |
3578 | return -EIO; | 3582 | return -EIO; |
3579 | 3583 | ||
3580 | return entry->store(q, page, length); | 3584 | return entry->store(q, page, length); |
3581 | } | 3585 | } |
3582 | 3586 | ||
3583 | static struct sysfs_ops queue_sysfs_ops = { | 3587 | static struct sysfs_ops queue_sysfs_ops = { |
3584 | .show = queue_attr_show, | 3588 | .show = queue_attr_show, |
3585 | .store = queue_attr_store, | 3589 | .store = queue_attr_store, |
3586 | }; | 3590 | }; |
3587 | 3591 | ||
3588 | static struct kobj_type queue_ktype = { | 3592 | static struct kobj_type queue_ktype = { |
3589 | .sysfs_ops = &queue_sysfs_ops, | 3593 | .sysfs_ops = &queue_sysfs_ops, |
3590 | .default_attrs = default_attrs, | 3594 | .default_attrs = default_attrs, |
3591 | }; | 3595 | }; |
3592 | 3596 | ||
3593 | int blk_register_queue(struct gendisk *disk) | 3597 | int blk_register_queue(struct gendisk *disk) |
3594 | { | 3598 | { |
3595 | int ret; | 3599 | int ret; |
3596 | 3600 | ||
3597 | request_queue_t *q = disk->queue; | 3601 | request_queue_t *q = disk->queue; |
3598 | 3602 | ||
3599 | if (!q || !q->request_fn) | 3603 | if (!q || !q->request_fn) |
3600 | return -ENXIO; | 3604 | return -ENXIO; |
3601 | 3605 | ||
3602 | q->kobj.parent = kobject_get(&disk->kobj); | 3606 | q->kobj.parent = kobject_get(&disk->kobj); |
3603 | if (!q->kobj.parent) | 3607 | if (!q->kobj.parent) |
3604 | return -EBUSY; | 3608 | return -EBUSY; |
3605 | 3609 | ||
3606 | snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); | 3610 | snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); |
3607 | q->kobj.ktype = &queue_ktype; | 3611 | q->kobj.ktype = &queue_ktype; |
3608 | 3612 | ||
3609 | ret = kobject_register(&q->kobj); | 3613 | ret = kobject_register(&q->kobj); |
3610 | if (ret < 0) | 3614 | if (ret < 0) |
3611 | return ret; | 3615 | return ret; |
3612 | 3616 | ||
3613 | ret = elv_register_queue(q); | 3617 | ret = elv_register_queue(q); |