Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit a038e2536472b4dd932399b5277e65f188811de5

Authored by Paolo 'Blaisorblade' Giarrusso
Committed by Jens Axboe
1 parent bae386f788

[PATCH] blk_start_queue() must be called with irq disabled - add warning 

The queue lock can be taken from interrupts so it must always be taken with
irq disabling primitives.  Some primitives already verify this.
blk_start_queue() is called under this lock, so interrupts must be
disabled.

Also document this requirement clearly in blk_init_queue(), where the queue
spinlock is set.

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Jens Axboe <axboe@suse.de>

Showing 1 changed file with 4 additions and 1 deletions Inline Diff

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/interrupt.h> 29 #include <linux/interrupt.h>
30 #include <linux/cpu.h> 30 #include <linux/cpu.h>
31 #include <linux/blktrace_api.h> 31 #include <linux/blktrace_api.h>
32 32
33 /* 33 /*
34 * for max sense size 34 * for max sense size
35 */ 35 */
36 #include <scsi/scsi_cmnd.h> 36 #include <scsi/scsi_cmnd.h>
37 37
38 static void blk_unplug_work(void *data); 38 static void blk_unplug_work(void *data);
39 static void blk_unplug_timeout(unsigned long data); 39 static void blk_unplug_timeout(unsigned long data);
40 static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io); 40 static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io);
41 static void init_request_from_bio(struct request *req, struct bio *bio); 41 static void init_request_from_bio(struct request *req, struct bio *bio);
42 static int __make_request(request_queue_t *q, struct bio *bio); 42 static int __make_request(request_queue_t *q, struct bio *bio);
43 43
44 /* 44 /*
45 * For the allocated request tables 45 * For the allocated request tables
46 */ 46 */
47 static kmem_cache_t *request_cachep; 47 static kmem_cache_t *request_cachep;
48 48
49 /* 49 /*
50 * For queue allocation 50 * For queue allocation
51 */ 51 */
52 static kmem_cache_t *requestq_cachep; 52 static kmem_cache_t *requestq_cachep;
53 53
54 /* 54 /*
55 * For io context allocations 55 * For io context allocations
56 */ 56 */
57 static kmem_cache_t *iocontext_cachep; 57 static kmem_cache_t *iocontext_cachep;
58 58
59 static wait_queue_head_t congestion_wqh[2] = { 59 static wait_queue_head_t congestion_wqh[2] = {
60 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), 60 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
61 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) 61 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
62 }; 62 };
63 63
64 /* 64 /*
65 * Controlling structure to kblockd 65 * Controlling structure to kblockd
66 */ 66 */
67 static struct workqueue_struct *kblockd_workqueue; 67 static struct workqueue_struct *kblockd_workqueue;
68 68
69 unsigned long blk_max_low_pfn, blk_max_pfn; 69 unsigned long blk_max_low_pfn, blk_max_pfn;
70 70
71 EXPORT_SYMBOL(blk_max_low_pfn); 71 EXPORT_SYMBOL(blk_max_low_pfn);
72 EXPORT_SYMBOL(blk_max_pfn); 72 EXPORT_SYMBOL(blk_max_pfn);
73 73
74 static DEFINE_PER_CPU(struct list_head, blk_cpu_done); 74 static DEFINE_PER_CPU(struct list_head, blk_cpu_done);
75 75
76 /* Amount of time in which a process may batch requests */ 76 /* Amount of time in which a process may batch requests */
77 #define BLK_BATCH_TIME (HZ/50UL) 77 #define BLK_BATCH_TIME (HZ/50UL)
78 78
79 /* Number of requests a "batching" process may submit */ 79 /* Number of requests a "batching" process may submit */
80 #define BLK_BATCH_REQ 32 80 #define BLK_BATCH_REQ 32
81 81
82 /* 82 /*
83 * Return the threshold (number of used requests) at which the queue is 83 * Return the threshold (number of used requests) at which the queue is
84 * considered to be congested. It include a little hysteresis to keep the 84 * considered to be congested. It include a little hysteresis to keep the
85 * context switch rate down. 85 * context switch rate down.
86 */ 86 */
87 static inline int queue_congestion_on_threshold(struct request_queue *q) 87 static inline int queue_congestion_on_threshold(struct request_queue *q)
88 { 88 {
89 return q->nr_congestion_on; 89 return q->nr_congestion_on;
90 } 90 }
91 91
92 /* 92 /*
93 * The threshold at which a queue is considered to be uncongested 93 * The threshold at which a queue is considered to be uncongested
94 */ 94 */
95 static inline int queue_congestion_off_threshold(struct request_queue *q) 95 static inline int queue_congestion_off_threshold(struct request_queue *q)
96 { 96 {
97 return q->nr_congestion_off; 97 return q->nr_congestion_off;
98 } 98 }
99 99
100 static void blk_queue_congestion_threshold(struct request_queue *q) 100 static void blk_queue_congestion_threshold(struct request_queue *q)
101 { 101 {
102 int nr; 102 int nr;
103 103
104 nr = q->nr_requests - (q->nr_requests / 8) + 1; 104 nr = q->nr_requests - (q->nr_requests / 8) + 1;
105 if (nr > q->nr_requests) 105 if (nr > q->nr_requests)
106 nr = q->nr_requests; 106 nr = q->nr_requests;
107 q->nr_congestion_on = nr; 107 q->nr_congestion_on = nr;
108 108
109 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; 109 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
110 if (nr < 1) 110 if (nr < 1)
111 nr = 1; 111 nr = 1;
112 q->nr_congestion_off = nr; 112 q->nr_congestion_off = nr;
113 } 113 }
114 114
115 /* 115 /*
116 * A queue has just exitted congestion. Note this in the global counter of 116 * A queue has just exitted congestion. Note this in the global counter of
117 * congested queues, and wake up anyone who was waiting for requests to be 117 * congested queues, and wake up anyone who was waiting for requests to be
118 * put back. 118 * put back.
119 */ 119 */
120 static void clear_queue_congested(request_queue_t *q, int rw) 120 static void clear_queue_congested(request_queue_t *q, int rw)
121 { 121 {
122 enum bdi_state bit; 122 enum bdi_state bit;
123 wait_queue_head_t *wqh = &congestion_wqh[rw]; 123 wait_queue_head_t *wqh = &congestion_wqh[rw];
124 124
125 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; 125 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
126 clear_bit(bit, &q->backing_dev_info.state); 126 clear_bit(bit, &q->backing_dev_info.state);
127 smp_mb__after_clear_bit(); 127 smp_mb__after_clear_bit();
128 if (waitqueue_active(wqh)) 128 if (waitqueue_active(wqh))
129 wake_up(wqh); 129 wake_up(wqh);
130 } 130 }
131 131
132 /* 132 /*
133 * A queue has just entered congestion. Flag that in the queue's VM-visible 133 * A queue has just entered congestion. Flag that in the queue's VM-visible
134 * state flags and increment the global gounter of congested queues. 134 * state flags and increment the global gounter of congested queues.
135 */ 135 */
136 static void set_queue_congested(request_queue_t *q, int rw) 136 static void set_queue_congested(request_queue_t *q, int rw)
137 { 137 {
138 enum bdi_state bit; 138 enum bdi_state bit;
139 139
140 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested; 140 bit = (rw == WRITE) ? BDI_write_congested : BDI_read_congested;
141 set_bit(bit, &q->backing_dev_info.state); 141 set_bit(bit, &q->backing_dev_info.state);
142 } 142 }
143 143
144 /** 144 /**
145 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info 145 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
146 * @bdev: device 146 * @bdev: device
147 * 147 *
148 * Locates the passed device's request queue and returns the address of its 148 * Locates the passed device's request queue and returns the address of its
149 * backing_dev_info 149 * backing_dev_info
150 * 150 *
151 * Will return NULL if the request queue cannot be located. 151 * Will return NULL if the request queue cannot be located.
152 */ 152 */
153 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) 153 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
154 { 154 {
155 struct backing_dev_info *ret = NULL; 155 struct backing_dev_info *ret = NULL;
156 request_queue_t *q = bdev_get_queue(bdev); 156 request_queue_t *q = bdev_get_queue(bdev);
157 157
158 if (q) 158 if (q)
159 ret = &q->backing_dev_info; 159 ret = &q->backing_dev_info;
160 return ret; 160 return ret;
161 } 161 }
162 162
163 EXPORT_SYMBOL(blk_get_backing_dev_info); 163 EXPORT_SYMBOL(blk_get_backing_dev_info);
164 164
165 void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data) 165 void blk_queue_activity_fn(request_queue_t *q, activity_fn *fn, void *data)
166 { 166 {
167 q->activity_fn = fn; 167 q->activity_fn = fn;
168 q->activity_data = data; 168 q->activity_data = data;
169 } 169 }
170 170
171 EXPORT_SYMBOL(blk_queue_activity_fn); 171 EXPORT_SYMBOL(blk_queue_activity_fn);
172 172
173 /** 173 /**
174 * blk_queue_prep_rq - set a prepare_request function for queue 174 * blk_queue_prep_rq - set a prepare_request function for queue
175 * @q: queue 175 * @q: queue
176 * @pfn: prepare_request function 176 * @pfn: prepare_request function
177 * 177 *
178 * It's possible for a queue to register a prepare_request callback which 178 * It's possible for a queue to register a prepare_request callback which
179 * is invoked before the request is handed to the request_fn. The goal of 179 * is invoked before the request is handed to the request_fn. The goal of
180 * the function is to prepare a request for I/O, it can be used to build a 180 * the function is to prepare a request for I/O, it can be used to build a
181 * cdb from the request data for instance. 181 * cdb from the request data for instance.
182 * 182 *
183 */ 183 */
184 void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn) 184 void blk_queue_prep_rq(request_queue_t *q, prep_rq_fn *pfn)
185 { 185 {
186 q->prep_rq_fn = pfn; 186 q->prep_rq_fn = pfn;
187 } 187 }
188 188
189 EXPORT_SYMBOL(blk_queue_prep_rq); 189 EXPORT_SYMBOL(blk_queue_prep_rq);
190 190
191 /** 191 /**
192 * blk_queue_merge_bvec - set a merge_bvec function for queue 192 * blk_queue_merge_bvec - set a merge_bvec function for queue
193 * @q: queue 193 * @q: queue
194 * @mbfn: merge_bvec_fn 194 * @mbfn: merge_bvec_fn
195 * 195 *
196 * Usually queues have static limitations on the max sectors or segments that 196 * Usually queues have static limitations on the max sectors or segments that
197 * we can put in a request. Stacking drivers may have some settings that 197 * we can put in a request. Stacking drivers may have some settings that
198 * are dynamic, and thus we have to query the queue whether it is ok to 198 * are dynamic, and thus we have to query the queue whether it is ok to
199 * add a new bio_vec to a bio at a given offset or not. If the block device 199 * add a new bio_vec to a bio at a given offset or not. If the block device
200 * has such limitations, it needs to register a merge_bvec_fn to control 200 * has such limitations, it needs to register a merge_bvec_fn to control
201 * the size of bio's sent to it. Note that a block device *must* allow a 201 * the size of bio's sent to it. Note that a block device *must* allow a
202 * single page to be added to an empty bio. The block device driver may want 202 * single page to be added to an empty bio. The block device driver may want
203 * to use the bio_split() function to deal with these bio's. By default 203 * to use the bio_split() function to deal with these bio's. By default
204 * no merge_bvec_fn is defined for a queue, and only the fixed limits are 204 * no merge_bvec_fn is defined for a queue, and only the fixed limits are
205 * honored. 205 * honored.
206 */ 206 */
207 void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn) 207 void blk_queue_merge_bvec(request_queue_t *q, merge_bvec_fn *mbfn)
208 { 208 {
209 q->merge_bvec_fn = mbfn; 209 q->merge_bvec_fn = mbfn;
210 } 210 }
211 211
212 EXPORT_SYMBOL(blk_queue_merge_bvec); 212 EXPORT_SYMBOL(blk_queue_merge_bvec);
213 213
214 void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn) 214 void blk_queue_softirq_done(request_queue_t *q, softirq_done_fn *fn)
215 { 215 {
216 q->softirq_done_fn = fn; 216 q->softirq_done_fn = fn;
217 } 217 }
218 218
219 EXPORT_SYMBOL(blk_queue_softirq_done); 219 EXPORT_SYMBOL(blk_queue_softirq_done);
220 220
221 /** 221 /**
222 * blk_queue_make_request - define an alternate make_request function for a device 222 * blk_queue_make_request - define an alternate make_request function for a device
223 * @q: the request queue for the device to be affected 223 * @q: the request queue for the device to be affected
224 * @mfn: the alternate make_request function 224 * @mfn: the alternate make_request function
225 * 225 *
226 * Description: 226 * Description:
227 * The normal way for &struct bios to be passed to a device 227 * The normal way for &struct bios to be passed to a device
228 * driver is for them to be collected into requests on a request 228 * driver is for them to be collected into requests on a request
229 * queue, and then to allow the device driver to select requests 229 * queue, and then to allow the device driver to select requests
230 * off that queue when it is ready. This works well for many block 230 * off that queue when it is ready. This works well for many block
231 * devices. However some block devices (typically virtual devices 231 * devices. However some block devices (typically virtual devices
232 * such as md or lvm) do not benefit from the processing on the 232 * such as md or lvm) do not benefit from the processing on the
233 * request queue, and are served best by having the requests passed 233 * request queue, and are served best by having the requests passed
234 * directly to them. This can be achieved by providing a function 234 * directly to them. This can be achieved by providing a function
235 * to blk_queue_make_request(). 235 * to blk_queue_make_request().
236 * 236 *
237 * Caveat: 237 * Caveat:
238 * The driver that does this *must* be able to deal appropriately 238 * The driver that does this *must* be able to deal appropriately
239 * with buffers in "highmemory". This can be accomplished by either calling 239 * with buffers in "highmemory". This can be accomplished by either calling
240 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling 240 * __bio_kmap_atomic() to get a temporary kernel mapping, or by calling
241 * blk_queue_bounce() to create a buffer in normal memory. 241 * blk_queue_bounce() to create a buffer in normal memory.
242 **/ 242 **/
243 void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn) 243 void blk_queue_make_request(request_queue_t * q, make_request_fn * mfn)
244 { 244 {
245 /* 245 /*
246 * set defaults 246 * set defaults
247 */ 247 */
248 q->nr_requests = BLKDEV_MAX_RQ; 248 q->nr_requests = BLKDEV_MAX_RQ;
249 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); 249 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
250 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); 250 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
251 q->make_request_fn = mfn; 251 q->make_request_fn = mfn;
252 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; 252 q->backing_dev_info.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
253 q->backing_dev_info.state = 0; 253 q->backing_dev_info.state = 0;
254 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; 254 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
255 blk_queue_max_sectors(q, SAFE_MAX_SECTORS); 255 blk_queue_max_sectors(q, SAFE_MAX_SECTORS);
256 blk_queue_hardsect_size(q, 512); 256 blk_queue_hardsect_size(q, 512);
257 blk_queue_dma_alignment(q, 511); 257 blk_queue_dma_alignment(q, 511);
258 blk_queue_congestion_threshold(q); 258 blk_queue_congestion_threshold(q);
259 q->nr_batching = BLK_BATCH_REQ; 259 q->nr_batching = BLK_BATCH_REQ;
260 260
261 q->unplug_thresh = 4; /* hmm */ 261 q->unplug_thresh = 4; /* hmm */
262 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */ 262 q->unplug_delay = (3 * HZ) / 1000; /* 3 milliseconds */
263 if (q->unplug_delay == 0) 263 if (q->unplug_delay == 0)
264 q->unplug_delay = 1; 264 q->unplug_delay = 1;
265 265
266 INIT_WORK(&q->unplug_work, blk_unplug_work, q); 266 INIT_WORK(&q->unplug_work, blk_unplug_work, q);
267 267
268 q->unplug_timer.function = blk_unplug_timeout; 268 q->unplug_timer.function = blk_unplug_timeout;
269 q->unplug_timer.data = (unsigned long)q; 269 q->unplug_timer.data = (unsigned long)q;
270 270
271 /* 271 /*
272 * by default assume old behaviour and bounce for any highmem page 272 * by default assume old behaviour and bounce for any highmem page
273 */ 273 */
274 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH); 274 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
275 275
276 blk_queue_activity_fn(q, NULL, NULL); 276 blk_queue_activity_fn(q, NULL, NULL);
277 } 277 }
278 278
279 EXPORT_SYMBOL(blk_queue_make_request); 279 EXPORT_SYMBOL(blk_queue_make_request);
280 280
281 static inline void rq_init(request_queue_t *q, struct request *rq) 281 static inline void rq_init(request_queue_t *q, struct request *rq)
282 { 282 {
283 INIT_LIST_HEAD(&rq->queuelist); 283 INIT_LIST_HEAD(&rq->queuelist);
284 INIT_LIST_HEAD(&rq->donelist); 284 INIT_LIST_HEAD(&rq->donelist);
285 285
286 rq->errors = 0; 286 rq->errors = 0;
287 rq->rq_status = RQ_ACTIVE; 287 rq->rq_status = RQ_ACTIVE;
288 rq->bio = rq->biotail = NULL; 288 rq->bio = rq->biotail = NULL;
289 rq->ioprio = 0; 289 rq->ioprio = 0;
290 rq->buffer = NULL; 290 rq->buffer = NULL;
291 rq->ref_count = 1; 291 rq->ref_count = 1;
292 rq->q = q; 292 rq->q = q;
293 rq->waiting = NULL; 293 rq->waiting = NULL;
294 rq->special = NULL; 294 rq->special = NULL;
295 rq->data_len = 0; 295 rq->data_len = 0;
296 rq->data = NULL; 296 rq->data = NULL;
297 rq->nr_phys_segments = 0; 297 rq->nr_phys_segments = 0;
298 rq->sense = NULL; 298 rq->sense = NULL;
299 rq->end_io = NULL; 299 rq->end_io = NULL;
300 rq->end_io_data = NULL; 300 rq->end_io_data = NULL;
301 rq->completion_data = NULL; 301 rq->completion_data = NULL;
302 } 302 }
303 303
304 /** 304 /**
305 * blk_queue_ordered - does this queue support ordered writes 305 * blk_queue_ordered - does this queue support ordered writes
306 * @q: the request queue 306 * @q: the request queue
307 * @ordered: one of QUEUE_ORDERED_* 307 * @ordered: one of QUEUE_ORDERED_*
308 * @prepare_flush_fn: rq setup helper for cache flush ordered writes 308 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
309 * 309 *
310 * Description: 310 * Description:
311 * For journalled file systems, doing ordered writes on a commit 311 * For journalled file systems, doing ordered writes on a commit
312 * block instead of explicitly doing wait_on_buffer (which is bad 312 * block instead of explicitly doing wait_on_buffer (which is bad
313 * for performance) can be a big win. Block drivers supporting this 313 * for performance) can be a big win. Block drivers supporting this
314 * feature should call this function and indicate so. 314 * feature should call this function and indicate so.
315 * 315 *
316 **/ 316 **/
317 int blk_queue_ordered(request_queue_t *q, unsigned ordered, 317 int blk_queue_ordered(request_queue_t *q, unsigned ordered,
318 prepare_flush_fn *prepare_flush_fn) 318 prepare_flush_fn *prepare_flush_fn)
319 { 319 {
320 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) && 320 if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
321 prepare_flush_fn == NULL) { 321 prepare_flush_fn == NULL) {
322 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n"); 322 printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
323 return -EINVAL; 323 return -EINVAL;
324 } 324 }
325 325
326 if (ordered != QUEUE_ORDERED_NONE && 326 if (ordered != QUEUE_ORDERED_NONE &&
327 ordered != QUEUE_ORDERED_DRAIN && 327 ordered != QUEUE_ORDERED_DRAIN &&
328 ordered != QUEUE_ORDERED_DRAIN_FLUSH && 328 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
329 ordered != QUEUE_ORDERED_DRAIN_FUA && 329 ordered != QUEUE_ORDERED_DRAIN_FUA &&
330 ordered != QUEUE_ORDERED_TAG && 330 ordered != QUEUE_ORDERED_TAG &&
331 ordered != QUEUE_ORDERED_TAG_FLUSH && 331 ordered != QUEUE_ORDERED_TAG_FLUSH &&
332 ordered != QUEUE_ORDERED_TAG_FUA) { 332 ordered != QUEUE_ORDERED_TAG_FUA) {
333 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered); 333 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
334 return -EINVAL; 334 return -EINVAL;
335 } 335 }
336 336
337 q->ordered = ordered; 337 q->ordered = ordered;
338 q->next_ordered = ordered; 338 q->next_ordered = ordered;
339 q->prepare_flush_fn = prepare_flush_fn; 339 q->prepare_flush_fn = prepare_flush_fn;
340 340
341 return 0; 341 return 0;
342 } 342 }
343 343
344 EXPORT_SYMBOL(blk_queue_ordered); 344 EXPORT_SYMBOL(blk_queue_ordered);
345 345
346 /** 346 /**
347 * blk_queue_issue_flush_fn - set function for issuing a flush 347 * blk_queue_issue_flush_fn - set function for issuing a flush
348 * @q: the request queue 348 * @q: the request queue
349 * @iff: the function to be called issuing the flush 349 * @iff: the function to be called issuing the flush
350 * 350 *
351 * Description: 351 * Description:
352 * If a driver supports issuing a flush command, the support is notified 352 * If a driver supports issuing a flush command, the support is notified
353 * to the block layer by defining it through this call. 353 * to the block layer by defining it through this call.
354 * 354 *
355 **/ 355 **/
356 void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff) 356 void blk_queue_issue_flush_fn(request_queue_t *q, issue_flush_fn *iff)
357 { 357 {
358 q->issue_flush_fn = iff; 358 q->issue_flush_fn = iff;
359 } 359 }
360 360
361 EXPORT_SYMBOL(blk_queue_issue_flush_fn); 361 EXPORT_SYMBOL(blk_queue_issue_flush_fn);
362 362
363 /* 363 /*
364 * Cache flushing for ordered writes handling 364 * Cache flushing for ordered writes handling
365 */ 365 */
366 inline unsigned blk_ordered_cur_seq(request_queue_t *q) 366 inline unsigned blk_ordered_cur_seq(request_queue_t *q)
367 { 367 {
368 if (!q->ordseq) 368 if (!q->ordseq)
369 return 0; 369 return 0;
370 return 1 << ffz(q->ordseq); 370 return 1 << ffz(q->ordseq);
371 } 371 }
372 372
373 unsigned blk_ordered_req_seq(struct request *rq) 373 unsigned blk_ordered_req_seq(struct request *rq)
374 { 374 {
375 request_queue_t *q = rq->q; 375 request_queue_t *q = rq->q;
376 376
377 BUG_ON(q->ordseq == 0); 377 BUG_ON(q->ordseq == 0);
378 378
379 if (rq == &q->pre_flush_rq) 379 if (rq == &q->pre_flush_rq)
380 return QUEUE_ORDSEQ_PREFLUSH; 380 return QUEUE_ORDSEQ_PREFLUSH;
381 if (rq == &q->bar_rq) 381 if (rq == &q->bar_rq)
382 return QUEUE_ORDSEQ_BAR; 382 return QUEUE_ORDSEQ_BAR;
383 if (rq == &q->post_flush_rq) 383 if (rq == &q->post_flush_rq)
384 return QUEUE_ORDSEQ_POSTFLUSH; 384 return QUEUE_ORDSEQ_POSTFLUSH;
385 385
386 if ((rq->flags & REQ_ORDERED_COLOR) == 386 if ((rq->flags & REQ_ORDERED_COLOR) ==
387 (q->orig_bar_rq->flags & REQ_ORDERED_COLOR)) 387 (q->orig_bar_rq->flags & REQ_ORDERED_COLOR))
388 return QUEUE_ORDSEQ_DRAIN; 388 return QUEUE_ORDSEQ_DRAIN;
389 else 389 else
390 return QUEUE_ORDSEQ_DONE; 390 return QUEUE_ORDSEQ_DONE;
391 } 391 }
392 392
393 void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error) 393 void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
394 { 394 {
395 struct request *rq; 395 struct request *rq;
396 int uptodate; 396 int uptodate;
397 397
398 if (error && !q->orderr) 398 if (error && !q->orderr)
399 q->orderr = error; 399 q->orderr = error;
400 400
401 BUG_ON(q->ordseq & seq); 401 BUG_ON(q->ordseq & seq);
402 q->ordseq |= seq; 402 q->ordseq |= seq;
403 403
404 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE) 404 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
405 return; 405 return;
406 406
407 /* 407 /*
408 * Okay, sequence complete. 408 * Okay, sequence complete.
409 */ 409 */
410 rq = q->orig_bar_rq; 410 rq = q->orig_bar_rq;
411 uptodate = q->orderr ? q->orderr : 1; 411 uptodate = q->orderr ? q->orderr : 1;
412 412
413 q->ordseq = 0; 413 q->ordseq = 0;
414 414
415 end_that_request_first(rq, uptodate, rq->hard_nr_sectors); 415 end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
416 end_that_request_last(rq, uptodate); 416 end_that_request_last(rq, uptodate);
417 } 417 }
418 418
419 static void pre_flush_end_io(struct request *rq, int error) 419 static void pre_flush_end_io(struct request *rq, int error)
420 { 420 {
421 elv_completed_request(rq->q, rq); 421 elv_completed_request(rq->q, rq);
422 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error); 422 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
423 } 423 }
424 424
425 static void bar_end_io(struct request *rq, int error) 425 static void bar_end_io(struct request *rq, int error)
426 { 426 {
427 elv_completed_request(rq->q, rq); 427 elv_completed_request(rq->q, rq);
428 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error); 428 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
429 } 429 }
430 430
431 static void post_flush_end_io(struct request *rq, int error) 431 static void post_flush_end_io(struct request *rq, int error)
432 { 432 {
433 elv_completed_request(rq->q, rq); 433 elv_completed_request(rq->q, rq);
434 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error); 434 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
435 } 435 }
436 436
437 static void queue_flush(request_queue_t *q, unsigned which) 437 static void queue_flush(request_queue_t *q, unsigned which)
438 { 438 {
439 struct request *rq; 439 struct request *rq;
440 rq_end_io_fn *end_io; 440 rq_end_io_fn *end_io;
441 441
442 if (which == QUEUE_ORDERED_PREFLUSH) { 442 if (which == QUEUE_ORDERED_PREFLUSH) {
443 rq = &q->pre_flush_rq; 443 rq = &q->pre_flush_rq;
444 end_io = pre_flush_end_io; 444 end_io = pre_flush_end_io;
445 } else { 445 } else {
446 rq = &q->post_flush_rq; 446 rq = &q->post_flush_rq;
447 end_io = post_flush_end_io; 447 end_io = post_flush_end_io;
448 } 448 }
449 449
450 rq_init(q, rq); 450 rq_init(q, rq);
451 rq->flags = REQ_HARDBARRIER; 451 rq->flags = REQ_HARDBARRIER;
452 rq->elevator_private = NULL; 452 rq->elevator_private = NULL;
453 rq->rq_disk = q->bar_rq.rq_disk; 453 rq->rq_disk = q->bar_rq.rq_disk;
454 rq->rl = NULL; 454 rq->rl = NULL;
455 rq->end_io = end_io; 455 rq->end_io = end_io;
456 q->prepare_flush_fn(q, rq); 456 q->prepare_flush_fn(q, rq);
457 457
458 elv_insert(q, rq, ELEVATOR_INSERT_FRONT); 458 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
459 } 459 }
460 460
461 static inline struct request *start_ordered(request_queue_t *q, 461 static inline struct request *start_ordered(request_queue_t *q,
462 struct request *rq) 462 struct request *rq)
463 { 463 {
464 q->bi_size = 0; 464 q->bi_size = 0;
465 q->orderr = 0; 465 q->orderr = 0;
466 q->ordered = q->next_ordered; 466 q->ordered = q->next_ordered;
467 q->ordseq |= QUEUE_ORDSEQ_STARTED; 467 q->ordseq |= QUEUE_ORDSEQ_STARTED;
468 468
469 /* 469 /*
470 * Prep proxy barrier request. 470 * Prep proxy barrier request.
471 */ 471 */
472 blkdev_dequeue_request(rq); 472 blkdev_dequeue_request(rq);
473 q->orig_bar_rq = rq; 473 q->orig_bar_rq = rq;
474 rq = &q->bar_rq; 474 rq = &q->bar_rq;
475 rq_init(q, rq); 475 rq_init(q, rq);
476 rq->flags = bio_data_dir(q->orig_bar_rq->bio); 476 rq->flags = bio_data_dir(q->orig_bar_rq->bio);
477 rq->flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0; 477 rq->flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
478 rq->elevator_private = NULL; 478 rq->elevator_private = NULL;
479 rq->rl = NULL; 479 rq->rl = NULL;
480 init_request_from_bio(rq, q->orig_bar_rq->bio); 480 init_request_from_bio(rq, q->orig_bar_rq->bio);
481 rq->end_io = bar_end_io; 481 rq->end_io = bar_end_io;
482 482
483 /* 483 /*
484 * Queue ordered sequence. As we stack them at the head, we 484 * Queue ordered sequence. As we stack them at the head, we
485 * need to queue in reverse order. Note that we rely on that 485 * need to queue in reverse order. Note that we rely on that
486 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs 486 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
487 * request gets inbetween ordered sequence. 487 * request gets inbetween ordered sequence.
488 */ 488 */
489 if (q->ordered & QUEUE_ORDERED_POSTFLUSH) 489 if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
490 queue_flush(q, QUEUE_ORDERED_POSTFLUSH); 490 queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
491 else 491 else
492 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH; 492 q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
493 493
494 elv_insert(q, rq, ELEVATOR_INSERT_FRONT); 494 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
495 495
496 if (q->ordered & QUEUE_ORDERED_PREFLUSH) { 496 if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
497 queue_flush(q, QUEUE_ORDERED_PREFLUSH); 497 queue_flush(q, QUEUE_ORDERED_PREFLUSH);
498 rq = &q->pre_flush_rq; 498 rq = &q->pre_flush_rq;
499 } else 499 } else
500 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH; 500 q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
501 501
502 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0) 502 if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
503 q->ordseq |= QUEUE_ORDSEQ_DRAIN; 503 q->ordseq |= QUEUE_ORDSEQ_DRAIN;
504 else 504 else
505 rq = NULL; 505 rq = NULL;
506 506
507 return rq; 507 return rq;
508 } 508 }
509 509
510 int blk_do_ordered(request_queue_t *q, struct request **rqp) 510 int blk_do_ordered(request_queue_t *q, struct request **rqp)
511 { 511 {
512 struct request *rq = *rqp; 512 struct request *rq = *rqp;
513 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq); 513 int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
514 514
515 if (!q->ordseq) { 515 if (!q->ordseq) {
516 if (!is_barrier) 516 if (!is_barrier)
517 return 1; 517 return 1;
518 518
519 if (q->next_ordered != QUEUE_ORDERED_NONE) { 519 if (q->next_ordered != QUEUE_ORDERED_NONE) {
520 *rqp = start_ordered(q, rq); 520 *rqp = start_ordered(q, rq);
521 return 1; 521 return 1;
522 } else { 522 } else {
523 /* 523 /*
524 * This can happen when the queue switches to 524 * This can happen when the queue switches to
525 * ORDERED_NONE while this request is on it. 525 * ORDERED_NONE while this request is on it.
526 */ 526 */
527 blkdev_dequeue_request(rq); 527 blkdev_dequeue_request(rq);
528 end_that_request_first(rq, -EOPNOTSUPP, 528 end_that_request_first(rq, -EOPNOTSUPP,
529 rq->hard_nr_sectors); 529 rq->hard_nr_sectors);
530 end_that_request_last(rq, -EOPNOTSUPP); 530 end_that_request_last(rq, -EOPNOTSUPP);
531 *rqp = NULL; 531 *rqp = NULL;
532 return 0; 532 return 0;
533 } 533 }
534 } 534 }
535 535
536 /* 536 /*
537 * Ordered sequence in progress 537 * Ordered sequence in progress
538 */ 538 */
539 539
540 /* Special requests are not subject to ordering rules. */ 540 /* Special requests are not subject to ordering rules. */
541 if (!blk_fs_request(rq) && 541 if (!blk_fs_request(rq) &&
542 rq != &q->pre_flush_rq && rq != &q->post_flush_rq) 542 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
543 return 1; 543 return 1;
544 544
545 if (q->ordered & QUEUE_ORDERED_TAG) { 545 if (q->ordered & QUEUE_ORDERED_TAG) {
546 /* Ordered by tag. Blocking the next barrier is enough. */ 546 /* Ordered by tag. Blocking the next barrier is enough. */
547 if (is_barrier && rq != &q->bar_rq) 547 if (is_barrier && rq != &q->bar_rq)
548 *rqp = NULL; 548 *rqp = NULL;
549 } else { 549 } else {
550 /* Ordered by draining. Wait for turn. */ 550 /* Ordered by draining. Wait for turn. */
551 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q)); 551 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
552 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q)) 552 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
553 *rqp = NULL; 553 *rqp = NULL;
554 } 554 }
555 555
556 return 1; 556 return 1;
557 } 557 }
558 558
559 static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error) 559 static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
560 { 560 {
561 request_queue_t *q = bio->bi_private; 561 request_queue_t *q = bio->bi_private;
562 struct bio_vec *bvec; 562 struct bio_vec *bvec;
563 int i; 563 int i;
564 564
565 /* 565 /*
566 * This is dry run, restore bio_sector and size. We'll finish 566 * This is dry run, restore bio_sector and size. We'll finish
567 * this request again with the original bi_end_io after an 567 * this request again with the original bi_end_io after an
568 * error occurs or post flush is complete. 568 * error occurs or post flush is complete.
569 */ 569 */
570 q->bi_size += bytes; 570 q->bi_size += bytes;
571 571
572 if (bio->bi_size) 572 if (bio->bi_size)
573 return 1; 573 return 1;
574 574
575 /* Rewind bvec's */ 575 /* Rewind bvec's */
576 bio->bi_idx = 0; 576 bio->bi_idx = 0;
577 bio_for_each_segment(bvec, bio, i) { 577 bio_for_each_segment(bvec, bio, i) {
578 bvec->bv_len += bvec->bv_offset; 578 bvec->bv_len += bvec->bv_offset;
579 bvec->bv_offset = 0; 579 bvec->bv_offset = 0;
580 } 580 }
581 581
582 /* Reset bio */ 582 /* Reset bio */
583 set_bit(BIO_UPTODATE, &bio->bi_flags); 583 set_bit(BIO_UPTODATE, &bio->bi_flags);
584 bio->bi_size = q->bi_size; 584 bio->bi_size = q->bi_size;
585 bio->bi_sector -= (q->bi_size >> 9); 585 bio->bi_sector -= (q->bi_size >> 9);
586 q->bi_size = 0; 586 q->bi_size = 0;
587 587
588 return 0; 588 return 0;
589 } 589 }
590 590
591 static inline int ordered_bio_endio(struct request *rq, struct bio *bio, 591 static inline int ordered_bio_endio(struct request *rq, struct bio *bio,
592 unsigned int nbytes, int error) 592 unsigned int nbytes, int error)
593 { 593 {
594 request_queue_t *q = rq->q; 594 request_queue_t *q = rq->q;
595 bio_end_io_t *endio; 595 bio_end_io_t *endio;
596 void *private; 596 void *private;
597 597
598 if (&q->bar_rq != rq) 598 if (&q->bar_rq != rq)
599 return 0; 599 return 0;
600 600
601 /* 601 /*
602 * Okay, this is the barrier request in progress, dry finish it. 602 * Okay, this is the barrier request in progress, dry finish it.
603 */ 603 */
604 if (error && !q->orderr) 604 if (error && !q->orderr)
605 q->orderr = error; 605 q->orderr = error;
606 606
607 endio = bio->bi_end_io; 607 endio = bio->bi_end_io;
608 private = bio->bi_private; 608 private = bio->bi_private;
609 bio->bi_end_io = flush_dry_bio_endio; 609 bio->bi_end_io = flush_dry_bio_endio;
610 bio->bi_private = q; 610 bio->bi_private = q;
611 611
612 bio_endio(bio, nbytes, error); 612 bio_endio(bio, nbytes, error);
613 613
614 bio->bi_end_io = endio; 614 bio->bi_end_io = endio;
615 bio->bi_private = private; 615 bio->bi_private = private;
616 616
617 return 1; 617 return 1;
618 } 618 }
619 619
620 /** 620 /**
621 * blk_queue_bounce_limit - set bounce buffer limit for queue 621 * blk_queue_bounce_limit - set bounce buffer limit for queue
622 * @q: the request queue for the device 622 * @q: the request queue for the device
623 * @dma_addr: bus address limit 623 * @dma_addr: bus address limit
624 * 624 *
625 * Description: 625 * Description:
626 * Different hardware can have different requirements as to what pages 626 * Different hardware can have different requirements as to what pages
627 * it can do I/O directly to. A low level driver can call 627 * it can do I/O directly to. A low level driver can call
628 * blk_queue_bounce_limit to have lower memory pages allocated as bounce 628 * blk_queue_bounce_limit to have lower memory pages allocated as bounce
629 * buffers for doing I/O to pages residing above @page. 629 * buffers for doing I/O to pages residing above @page.
630 **/ 630 **/
631 void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr) 631 void blk_queue_bounce_limit(request_queue_t *q, u64 dma_addr)
632 { 632 {
633 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT; 633 unsigned long bounce_pfn = dma_addr >> PAGE_SHIFT;
634 int dma = 0; 634 int dma = 0;
635 635
636 q->bounce_gfp = GFP_NOIO; 636 q->bounce_gfp = GFP_NOIO;
637 #if BITS_PER_LONG == 64 637 #if BITS_PER_LONG == 64
638 /* Assume anything <= 4GB can be handled by IOMMU. 638 /* Assume anything <= 4GB can be handled by IOMMU.
639 Actually some IOMMUs can handle everything, but I don't 639 Actually some IOMMUs can handle everything, but I don't
640 know of a way to test this here. */ 640 know of a way to test this here. */
641 if (bounce_pfn < (0xffffffff>>PAGE_SHIFT)) 641 if (bounce_pfn < (0xffffffff>>PAGE_SHIFT))
642 dma = 1; 642 dma = 1;
643 q->bounce_pfn = max_low_pfn; 643 q->bounce_pfn = max_low_pfn;
644 #else 644 #else
645 if (bounce_pfn < blk_max_low_pfn) 645 if (bounce_pfn < blk_max_low_pfn)
646 dma = 1; 646 dma = 1;
647 q->bounce_pfn = bounce_pfn; 647 q->bounce_pfn = bounce_pfn;
648 #endif 648 #endif
649 if (dma) { 649 if (dma) {
650 init_emergency_isa_pool(); 650 init_emergency_isa_pool();
651 q->bounce_gfp = GFP_NOIO | GFP_DMA; 651 q->bounce_gfp = GFP_NOIO | GFP_DMA;
652 q->bounce_pfn = bounce_pfn; 652 q->bounce_pfn = bounce_pfn;
653 } 653 }
654 } 654 }
655 655
656 EXPORT_SYMBOL(blk_queue_bounce_limit); 656 EXPORT_SYMBOL(blk_queue_bounce_limit);
657 657
658 /** 658 /**
659 * blk_queue_max_sectors - set max sectors for a request for this queue 659 * blk_queue_max_sectors - set max sectors for a request for this queue
660 * @q: the request queue for the device 660 * @q: the request queue for the device
661 * @max_sectors: max sectors in the usual 512b unit 661 * @max_sectors: max sectors in the usual 512b unit
662 * 662 *
663 * Description: 663 * Description:
664 * Enables a low level driver to set an upper limit on the size of 664 * Enables a low level driver to set an upper limit on the size of
665 * received requests. 665 * received requests.
666 **/ 666 **/
667 void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors) 667 void blk_queue_max_sectors(request_queue_t *q, unsigned int max_sectors)
668 { 668 {
669 if ((max_sectors << 9) < PAGE_CACHE_SIZE) { 669 if ((max_sectors << 9) < PAGE_CACHE_SIZE) {
670 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9); 670 max_sectors = 1 << (PAGE_CACHE_SHIFT - 9);
671 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors); 671 printk("%s: set to minimum %d\n", __FUNCTION__, max_sectors);
672 } 672 }
673 673
674 if (BLK_DEF_MAX_SECTORS > max_sectors) 674 if (BLK_DEF_MAX_SECTORS > max_sectors)
675 q->max_hw_sectors = q->max_sectors = max_sectors; 675 q->max_hw_sectors = q->max_sectors = max_sectors;
676 else { 676 else {
677 q->max_sectors = BLK_DEF_MAX_SECTORS; 677 q->max_sectors = BLK_DEF_MAX_SECTORS;
678 q->max_hw_sectors = max_sectors; 678 q->max_hw_sectors = max_sectors;
679 } 679 }
680 } 680 }
681 681
682 EXPORT_SYMBOL(blk_queue_max_sectors); 682 EXPORT_SYMBOL(blk_queue_max_sectors);
683 683
684 /** 684 /**
685 * blk_queue_max_phys_segments - set max phys segments for a request for this queue 685 * blk_queue_max_phys_segments - set max phys segments for a request for this queue
686 * @q: the request queue for the device 686 * @q: the request queue for the device
687 * @max_segments: max number of segments 687 * @max_segments: max number of segments
688 * 688 *
689 * Description: 689 * Description:
690 * Enables a low level driver to set an upper limit on the number of 690 * Enables a low level driver to set an upper limit on the number of
691 * physical data segments in a request. This would be the largest sized 691 * physical data segments in a request. This would be the largest sized
692 * scatter list the driver could handle. 692 * scatter list the driver could handle.
693 **/ 693 **/
694 void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments) 694 void blk_queue_max_phys_segments(request_queue_t *q, unsigned short max_segments)
695 { 695 {
696 if (!max_segments) { 696 if (!max_segments) {
697 max_segments = 1; 697 max_segments = 1;
698 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); 698 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
699 } 699 }
700 700
701 q->max_phys_segments = max_segments; 701 q->max_phys_segments = max_segments;
702 } 702 }
703 703
704 EXPORT_SYMBOL(blk_queue_max_phys_segments); 704 EXPORT_SYMBOL(blk_queue_max_phys_segments);
705 705
706 /** 706 /**
707 * blk_queue_max_hw_segments - set max hw segments for a request for this queue 707 * blk_queue_max_hw_segments - set max hw segments for a request for this queue
708 * @q: the request queue for the device 708 * @q: the request queue for the device
709 * @max_segments: max number of segments 709 * @max_segments: max number of segments
710 * 710 *
711 * Description: 711 * Description:
712 * Enables a low level driver to set an upper limit on the number of 712 * Enables a low level driver to set an upper limit on the number of
713 * hw data segments in a request. This would be the largest number of 713 * hw data segments in a request. This would be the largest number of
714 * address/length pairs the host adapter can actually give as once 714 * address/length pairs the host adapter can actually give as once
715 * to the device. 715 * to the device.
716 **/ 716 **/
717 void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments) 717 void blk_queue_max_hw_segments(request_queue_t *q, unsigned short max_segments)
718 { 718 {
719 if (!max_segments) { 719 if (!max_segments) {
720 max_segments = 1; 720 max_segments = 1;
721 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments); 721 printk("%s: set to minimum %d\n", __FUNCTION__, max_segments);
722 } 722 }
723 723
724 q->max_hw_segments = max_segments; 724 q->max_hw_segments = max_segments;
725 } 725 }
726 726
727 EXPORT_SYMBOL(blk_queue_max_hw_segments); 727 EXPORT_SYMBOL(blk_queue_max_hw_segments);
728 728
729 /** 729 /**
730 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg 730 * blk_queue_max_segment_size - set max segment size for blk_rq_map_sg
731 * @q: the request queue for the device 731 * @q: the request queue for the device
732 * @max_size: max size of segment in bytes 732 * @max_size: max size of segment in bytes
733 * 733 *
734 * Description: 734 * Description:
735 * Enables a low level driver to set an upper limit on the size of a 735 * Enables a low level driver to set an upper limit on the size of a
736 * coalesced segment 736 * coalesced segment
737 **/ 737 **/
738 void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size) 738 void blk_queue_max_segment_size(request_queue_t *q, unsigned int max_size)
739 { 739 {
740 if (max_size < PAGE_CACHE_SIZE) { 740 if (max_size < PAGE_CACHE_SIZE) {
741 max_size = PAGE_CACHE_SIZE; 741 max_size = PAGE_CACHE_SIZE;
742 printk("%s: set to minimum %d\n", __FUNCTION__, max_size); 742 printk("%s: set to minimum %d\n", __FUNCTION__, max_size);
743 } 743 }
744 744
745 q->max_segment_size = max_size; 745 q->max_segment_size = max_size;
746 } 746 }
747 747
748 EXPORT_SYMBOL(blk_queue_max_segment_size); 748 EXPORT_SYMBOL(blk_queue_max_segment_size);
749 749
750 /** 750 /**
751 * blk_queue_hardsect_size - set hardware sector size for the queue 751 * blk_queue_hardsect_size - set hardware sector size for the queue
752 * @q: the request queue for the device 752 * @q: the request queue for the device
753 * @size: the hardware sector size, in bytes 753 * @size: the hardware sector size, in bytes
754 * 754 *
755 * Description: 755 * Description:
756 * This should typically be set to the lowest possible sector size 756 * This should typically be set to the lowest possible sector size
757 * that the hardware can operate on (possible without reverting to 757 * that the hardware can operate on (possible without reverting to
758 * even internal read-modify-write operations). Usually the default 758 * even internal read-modify-write operations). Usually the default
759 * of 512 covers most hardware. 759 * of 512 covers most hardware.
760 **/ 760 **/
761 void blk_queue_hardsect_size(request_queue_t *q, unsigned short size) 761 void blk_queue_hardsect_size(request_queue_t *q, unsigned short size)
762 { 762 {
763 q->hardsect_size = size; 763 q->hardsect_size = size;
764 } 764 }
765 765
766 EXPORT_SYMBOL(blk_queue_hardsect_size); 766 EXPORT_SYMBOL(blk_queue_hardsect_size);
767 767
768 /* 768 /*
769 * Returns the minimum that is _not_ zero, unless both are zero. 769 * Returns the minimum that is _not_ zero, unless both are zero.
770 */ 770 */
771 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r)) 771 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
772 772
773 /** 773 /**
774 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers 774 * blk_queue_stack_limits - inherit underlying queue limits for stacked drivers
775 * @t: the stacking driver (top) 775 * @t: the stacking driver (top)
776 * @b: the underlying device (bottom) 776 * @b: the underlying device (bottom)
777 **/ 777 **/
778 void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b) 778 void blk_queue_stack_limits(request_queue_t *t, request_queue_t *b)
779 { 779 {
780 /* zero is "infinity" */ 780 /* zero is "infinity" */
781 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors); 781 t->max_sectors = min_not_zero(t->max_sectors,b->max_sectors);
782 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors); 782 t->max_hw_sectors = min_not_zero(t->max_hw_sectors,b->max_hw_sectors);
783 783
784 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments); 784 t->max_phys_segments = min(t->max_phys_segments,b->max_phys_segments);
785 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments); 785 t->max_hw_segments = min(t->max_hw_segments,b->max_hw_segments);
786 t->max_segment_size = min(t->max_segment_size,b->max_segment_size); 786 t->max_segment_size = min(t->max_segment_size,b->max_segment_size);
787 t->hardsect_size = max(t->hardsect_size,b->hardsect_size); 787 t->hardsect_size = max(t->hardsect_size,b->hardsect_size);
788 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) 788 if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags))
789 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags); 789 clear_bit(QUEUE_FLAG_CLUSTER, &t->queue_flags);
790 } 790 }
791 791
792 EXPORT_SYMBOL(blk_queue_stack_limits); 792 EXPORT_SYMBOL(blk_queue_stack_limits);
793 793
794 /** 794 /**
795 * blk_queue_segment_boundary - set boundary rules for segment merging 795 * blk_queue_segment_boundary - set boundary rules for segment merging
796 * @q: the request queue for the device 796 * @q: the request queue for the device
797 * @mask: the memory boundary mask 797 * @mask: the memory boundary mask
798 **/ 798 **/
799 void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask) 799 void blk_queue_segment_boundary(request_queue_t *q, unsigned long mask)
800 { 800 {
801 if (mask < PAGE_CACHE_SIZE - 1) { 801 if (mask < PAGE_CACHE_SIZE - 1) {
802 mask = PAGE_CACHE_SIZE - 1; 802 mask = PAGE_CACHE_SIZE - 1;
803 printk("%s: set to minimum %lx\n", __FUNCTION__, mask); 803 printk("%s: set to minimum %lx\n", __FUNCTION__, mask);
804 } 804 }
805 805
806 q->seg_boundary_mask = mask; 806 q->seg_boundary_mask = mask;
807 } 807 }
808 808
809 EXPORT_SYMBOL(blk_queue_segment_boundary); 809 EXPORT_SYMBOL(blk_queue_segment_boundary);
810 810
811 /** 811 /**
812 * blk_queue_dma_alignment - set dma length and memory alignment 812 * blk_queue_dma_alignment - set dma length and memory alignment
813 * @q: the request queue for the device 813 * @q: the request queue for the device
814 * @mask: alignment mask 814 * @mask: alignment mask
815 * 815 *
816 * description: 816 * description:
817 * set required memory and length aligment for direct dma transactions. 817 * set required memory and length aligment for direct dma transactions.
818 * this is used when buiding direct io requests for the queue. 818 * this is used when buiding direct io requests for the queue.
819 * 819 *
820 **/ 820 **/
821 void blk_queue_dma_alignment(request_queue_t *q, int mask) 821 void blk_queue_dma_alignment(request_queue_t *q, int mask)
822 { 822 {
823 q->dma_alignment = mask; 823 q->dma_alignment = mask;
824 } 824 }
825 825
826 EXPORT_SYMBOL(blk_queue_dma_alignment); 826 EXPORT_SYMBOL(blk_queue_dma_alignment);
827 827
828 /** 828 /**
829 * blk_queue_find_tag - find a request by its tag and queue 829 * blk_queue_find_tag - find a request by its tag and queue
830 * @q: The request queue for the device 830 * @q: The request queue for the device
831 * @tag: The tag of the request 831 * @tag: The tag of the request
832 * 832 *
833 * Notes: 833 * Notes:
834 * Should be used when a device returns a tag and you want to match 834 * Should be used when a device returns a tag and you want to match
835 * it with a request. 835 * it with a request.
836 * 836 *
837 * no locks need be held. 837 * no locks need be held.
838 **/ 838 **/
839 struct request *blk_queue_find_tag(request_queue_t *q, int tag) 839 struct request *blk_queue_find_tag(request_queue_t *q, int tag)
840 { 840 {
841 struct blk_queue_tag *bqt = q->queue_tags; 841 struct blk_queue_tag *bqt = q->queue_tags;
842 842
843 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth)) 843 if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
844 return NULL; 844 return NULL;
845 845
846 return bqt->tag_index[tag]; 846 return bqt->tag_index[tag];
847 } 847 }
848 848
849 EXPORT_SYMBOL(blk_queue_find_tag); 849 EXPORT_SYMBOL(blk_queue_find_tag);
850 850
851 /** 851 /**
852 * __blk_queue_free_tags - release tag maintenance info 852 * __blk_queue_free_tags - release tag maintenance info
853 * @q: the request queue for the device 853 * @q: the request queue for the device
854 * 854 *
855 * Notes: 855 * Notes:
856 * blk_cleanup_queue() will take care of calling this function, if tagging 856 * blk_cleanup_queue() will take care of calling this function, if tagging
857 * has been used. So there's no need to call this directly. 857 * has been used. So there's no need to call this directly.
858 **/ 858 **/
859 static void __blk_queue_free_tags(request_queue_t *q) 859 static void __blk_queue_free_tags(request_queue_t *q)
860 { 860 {
861 struct blk_queue_tag *bqt = q->queue_tags; 861 struct blk_queue_tag *bqt = q->queue_tags;
862 862
863 if (!bqt) 863 if (!bqt)
864 return; 864 return;
865 865
866 if (atomic_dec_and_test(&bqt->refcnt)) { 866 if (atomic_dec_and_test(&bqt->refcnt)) {
867 BUG_ON(bqt->busy); 867 BUG_ON(bqt->busy);
868 BUG_ON(!list_empty(&bqt->busy_list)); 868 BUG_ON(!list_empty(&bqt->busy_list));
869 869
870 kfree(bqt->tag_index); 870 kfree(bqt->tag_index);
871 bqt->tag_index = NULL; 871 bqt->tag_index = NULL;
872 872
873 kfree(bqt->tag_map); 873 kfree(bqt->tag_map);
874 bqt->tag_map = NULL; 874 bqt->tag_map = NULL;
875 875
876 kfree(bqt); 876 kfree(bqt);
877 } 877 }
878 878
879 q->queue_tags = NULL; 879 q->queue_tags = NULL;
880 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED); 880 q->queue_flags &= ~(1 << QUEUE_FLAG_QUEUED);
881 } 881 }
882 882
883 /** 883 /**
884 * blk_queue_free_tags - release tag maintenance info 884 * blk_queue_free_tags - release tag maintenance info
885 * @q: the request queue for the device 885 * @q: the request queue for the device
886 * 886 *
887 * Notes: 887 * Notes:
888 * This is used to disabled tagged queuing to a device, yet leave 888 * This is used to disabled tagged queuing to a device, yet leave
889 * queue in function. 889 * queue in function.
890 **/ 890 **/
891 void blk_queue_free_tags(request_queue_t *q) 891 void blk_queue_free_tags(request_queue_t *q)
892 { 892 {
893 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); 893 clear_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
894 } 894 }
895 895
896 EXPORT_SYMBOL(blk_queue_free_tags); 896 EXPORT_SYMBOL(blk_queue_free_tags);
897 897
898 static int 898 static int
899 init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth) 899 init_tag_map(request_queue_t *q, struct blk_queue_tag *tags, int depth)
900 { 900 {
901 struct request **tag_index; 901 struct request **tag_index;
902 unsigned long *tag_map; 902 unsigned long *tag_map;
903 int nr_ulongs; 903 int nr_ulongs;
904 904
905 if (depth > q->nr_requests * 2) { 905 if (depth > q->nr_requests * 2) {
906 depth = q->nr_requests * 2; 906 depth = q->nr_requests * 2;
907 printk(KERN_ERR "%s: adjusted depth to %d\n", 907 printk(KERN_ERR "%s: adjusted depth to %d\n",
908 __FUNCTION__, depth); 908 __FUNCTION__, depth);
909 } 909 }
910 910
911 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC); 911 tag_index = kzalloc(depth * sizeof(struct request *), GFP_ATOMIC);
912 if (!tag_index) 912 if (!tag_index)
913 goto fail; 913 goto fail;
914 914
915 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG; 915 nr_ulongs = ALIGN(depth, BITS_PER_LONG) / BITS_PER_LONG;
916 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC); 916 tag_map = kzalloc(nr_ulongs * sizeof(unsigned long), GFP_ATOMIC);
917 if (!tag_map) 917 if (!tag_map)
918 goto fail; 918 goto fail;
919 919
920 tags->real_max_depth = depth; 920 tags->real_max_depth = depth;
921 tags->max_depth = depth; 921 tags->max_depth = depth;
922 tags->tag_index = tag_index; 922 tags->tag_index = tag_index;
923 tags->tag_map = tag_map; 923 tags->tag_map = tag_map;
924 924
925 return 0; 925 return 0;
926 fail: 926 fail:
927 kfree(tag_index); 927 kfree(tag_index);
928 return -ENOMEM; 928 return -ENOMEM;
929 } 929 }
930 930
931 /** 931 /**
932 * blk_queue_init_tags - initialize the queue tag info 932 * blk_queue_init_tags - initialize the queue tag info
933 * @q: the request queue for the device 933 * @q: the request queue for the device
934 * @depth: the maximum queue depth supported 934 * @depth: the maximum queue depth supported
935 * @tags: the tag to use 935 * @tags: the tag to use
936 **/ 936 **/
937 int blk_queue_init_tags(request_queue_t *q, int depth, 937 int blk_queue_init_tags(request_queue_t *q, int depth,
938 struct blk_queue_tag *tags) 938 struct blk_queue_tag *tags)
939 { 939 {
940 int rc; 940 int rc;
941 941
942 BUG_ON(tags && q->queue_tags && tags != q->queue_tags); 942 BUG_ON(tags && q->queue_tags && tags != q->queue_tags);
943 943
944 if (!tags && !q->queue_tags) { 944 if (!tags && !q->queue_tags) {
945 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC); 945 tags = kmalloc(sizeof(struct blk_queue_tag), GFP_ATOMIC);
946 if (!tags) 946 if (!tags)
947 goto fail; 947 goto fail;
948 948
949 if (init_tag_map(q, tags, depth)) 949 if (init_tag_map(q, tags, depth))
950 goto fail; 950 goto fail;
951 951
952 INIT_LIST_HEAD(&tags->busy_list); 952 INIT_LIST_HEAD(&tags->busy_list);
953 tags->busy = 0; 953 tags->busy = 0;
954 atomic_set(&tags->refcnt, 1); 954 atomic_set(&tags->refcnt, 1);
955 } else if (q->queue_tags) { 955 } else if (q->queue_tags) {
956 if ((rc = blk_queue_resize_tags(q, depth))) 956 if ((rc = blk_queue_resize_tags(q, depth)))
957 return rc; 957 return rc;
958 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags); 958 set_bit(QUEUE_FLAG_QUEUED, &q->queue_flags);
959 return 0; 959 return 0;
960 } else 960 } else
961 atomic_inc(&tags->refcnt); 961 atomic_inc(&tags->refcnt);
962 962
963 /* 963 /*
964 * assign it, all done 964 * assign it, all done
965 */ 965 */
966 q->queue_tags = tags; 966 q->queue_tags = tags;
967 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED); 967 q->queue_flags |= (1 << QUEUE_FLAG_QUEUED);
968 return 0; 968 return 0;
969 fail: 969 fail:
970 kfree(tags); 970 kfree(tags);
971 return -ENOMEM; 971 return -ENOMEM;
972 } 972 }
973 973
974 EXPORT_SYMBOL(blk_queue_init_tags); 974 EXPORT_SYMBOL(blk_queue_init_tags);
975 975
976 /** 976 /**
977 * blk_queue_resize_tags - change the queueing depth 977 * blk_queue_resize_tags - change the queueing depth
978 * @q: the request queue for the device 978 * @q: the request queue for the device
979 * @new_depth: the new max command queueing depth 979 * @new_depth: the new max command queueing depth
980 * 980 *
981 * Notes: 981 * Notes:
982 * Must be called with the queue lock held. 982 * Must be called with the queue lock held.
983 **/ 983 **/
984 int blk_queue_resize_tags(request_queue_t *q, int new_depth) 984 int blk_queue_resize_tags(request_queue_t *q, int new_depth)
985 { 985 {
986 struct blk_queue_tag *bqt = q->queue_tags; 986 struct blk_queue_tag *bqt = q->queue_tags;
987 struct request **tag_index; 987 struct request **tag_index;
988 unsigned long *tag_map; 988 unsigned long *tag_map;
989 int max_depth, nr_ulongs; 989 int max_depth, nr_ulongs;
990 990
991 if (!bqt) 991 if (!bqt)
992 return -ENXIO; 992 return -ENXIO;
993 993
994 /* 994 /*
995 * if we already have large enough real_max_depth. just 995 * if we already have large enough real_max_depth. just
996 * adjust max_depth. *NOTE* as requests with tag value 996 * adjust max_depth. *NOTE* as requests with tag value
997 * between new_depth and real_max_depth can be in-flight, tag 997 * between new_depth and real_max_depth can be in-flight, tag
998 * map can not be shrunk blindly here. 998 * map can not be shrunk blindly here.
999 */ 999 */
1000 if (new_depth <= bqt->real_max_depth) { 1000 if (new_depth <= bqt->real_max_depth) {
1001 bqt->max_depth = new_depth; 1001 bqt->max_depth = new_depth;
1002 return 0; 1002 return 0;
1003 } 1003 }
1004 1004
1005 /* 1005 /*
1006 * save the old state info, so we can copy it back 1006 * save the old state info, so we can copy it back
1007 */ 1007 */
1008 tag_index = bqt->tag_index; 1008 tag_index = bqt->tag_index;
1009 tag_map = bqt->tag_map; 1009 tag_map = bqt->tag_map;
1010 max_depth = bqt->real_max_depth; 1010 max_depth = bqt->real_max_depth;
1011 1011
1012 if (init_tag_map(q, bqt, new_depth)) 1012 if (init_tag_map(q, bqt, new_depth))
1013 return -ENOMEM; 1013 return -ENOMEM;
1014 1014
1015 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *)); 1015 memcpy(bqt->tag_index, tag_index, max_depth * sizeof(struct request *));
1016 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG; 1016 nr_ulongs = ALIGN(max_depth, BITS_PER_LONG) / BITS_PER_LONG;
1017 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long)); 1017 memcpy(bqt->tag_map, tag_map, nr_ulongs * sizeof(unsigned long));
1018 1018
1019 kfree(tag_index); 1019 kfree(tag_index);
1020 kfree(tag_map); 1020 kfree(tag_map);
1021 return 0; 1021 return 0;
1022 } 1022 }
1023 1023
1024 EXPORT_SYMBOL(blk_queue_resize_tags); 1024 EXPORT_SYMBOL(blk_queue_resize_tags);
1025 1025
1026 /** 1026 /**
1027 * blk_queue_end_tag - end tag operations for a request 1027 * blk_queue_end_tag - end tag operations for a request
1028 * @q: the request queue for the device 1028 * @q: the request queue for the device
1029 * @rq: the request that has completed 1029 * @rq: the request that has completed
1030 * 1030 *
1031 * Description: 1031 * Description:
1032 * Typically called when end_that_request_first() returns 0, meaning 1032 * Typically called when end_that_request_first() returns 0, meaning
1033 * all transfers have been done for a request. It's important to call 1033 * all transfers have been done for a request. It's important to call
1034 * this function before end_that_request_last(), as that will put the 1034 * this function before end_that_request_last(), as that will put the
1035 * request back on the free list thus corrupting the internal tag list. 1035 * request back on the free list thus corrupting the internal tag list.
1036 * 1036 *
1037 * Notes: 1037 * Notes:
1038 * queue lock must be held. 1038 * queue lock must be held.
1039 **/ 1039 **/
1040 void blk_queue_end_tag(request_queue_t *q, struct request *rq) 1040 void blk_queue_end_tag(request_queue_t *q, struct request *rq)
1041 { 1041 {
1042 struct blk_queue_tag *bqt = q->queue_tags; 1042 struct blk_queue_tag *bqt = q->queue_tags;
1043 int tag = rq->tag; 1043 int tag = rq->tag;
1044 1044
1045 BUG_ON(tag == -1); 1045 BUG_ON(tag == -1);
1046 1046
1047 if (unlikely(tag >= bqt->real_max_depth)) 1047 if (unlikely(tag >= bqt->real_max_depth))
1048 /* 1048 /*
1049 * This can happen after tag depth has been reduced. 1049 * This can happen after tag depth has been reduced.
1050 * FIXME: how about a warning or info message here? 1050 * FIXME: how about a warning or info message here?
1051 */ 1051 */
1052 return; 1052 return;
1053 1053
1054 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) { 1054 if (unlikely(!__test_and_clear_bit(tag, bqt->tag_map))) {
1055 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n", 1055 printk(KERN_ERR "%s: attempt to clear non-busy tag (%d)\n",
1056 __FUNCTION__, tag); 1056 __FUNCTION__, tag);
1057 return; 1057 return;
1058 } 1058 }
1059 1059
1060 list_del_init(&rq->queuelist); 1060 list_del_init(&rq->queuelist);
1061 rq->flags &= ~REQ_QUEUED; 1061 rq->flags &= ~REQ_QUEUED;
1062 rq->tag = -1; 1062 rq->tag = -1;
1063 1063
1064 if (unlikely(bqt->tag_index[tag] == NULL)) 1064 if (unlikely(bqt->tag_index[tag] == NULL))
1065 printk(KERN_ERR "%s: tag %d is missing\n", 1065 printk(KERN_ERR "%s: tag %d is missing\n",
1066 __FUNCTION__, tag); 1066 __FUNCTION__, tag);
1067 1067
1068 bqt->tag_index[tag] = NULL; 1068 bqt->tag_index[tag] = NULL;
1069 bqt->busy--; 1069 bqt->busy--;
1070 } 1070 }
1071 1071
1072 EXPORT_SYMBOL(blk_queue_end_tag); 1072 EXPORT_SYMBOL(blk_queue_end_tag);
1073 1073
1074 /** 1074 /**
1075 * blk_queue_start_tag - find a free tag and assign it 1075 * blk_queue_start_tag - find a free tag and assign it
1076 * @q: the request queue for the device 1076 * @q: the request queue for the device
1077 * @rq: the block request that needs tagging 1077 * @rq: the block request that needs tagging
1078 * 1078 *
1079 * Description: 1079 * Description:
1080 * This can either be used as a stand-alone helper, or possibly be 1080 * This can either be used as a stand-alone helper, or possibly be
1081 * assigned as the queue &prep_rq_fn (in which case &struct request 1081 * assigned as the queue &prep_rq_fn (in which case &struct request
1082 * automagically gets a tag assigned). Note that this function 1082 * automagically gets a tag assigned). Note that this function
1083 * assumes that any type of request can be queued! if this is not 1083 * assumes that any type of request can be queued! if this is not
1084 * true for your device, you must check the request type before 1084 * true for your device, you must check the request type before
1085 * calling this function. The request will also be removed from 1085 * calling this function. The request will also be removed from
1086 * the request queue, so it's the drivers responsibility to readd 1086 * the request queue, so it's the drivers responsibility to readd
1087 * it if it should need to be restarted for some reason. 1087 * it if it should need to be restarted for some reason.
1088 * 1088 *
1089 * Notes: 1089 * Notes:
1090 * queue lock must be held. 1090 * queue lock must be held.
1091 **/ 1091 **/
1092 int blk_queue_start_tag(request_queue_t *q, struct request *rq) 1092 int blk_queue_start_tag(request_queue_t *q, struct request *rq)
1093 { 1093 {
1094 struct blk_queue_tag *bqt = q->queue_tags; 1094 struct blk_queue_tag *bqt = q->queue_tags;
1095 int tag; 1095 int tag;
1096 1096
1097 if (unlikely((rq->flags & REQ_QUEUED))) { 1097 if (unlikely((rq->flags & REQ_QUEUED))) {
1098 printk(KERN_ERR 1098 printk(KERN_ERR
1099 "%s: request %p for device [%s] already tagged %d", 1099 "%s: request %p for device [%s] already tagged %d",
1100 __FUNCTION__, rq, 1100 __FUNCTION__, rq,
1101 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag); 1101 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->tag);
1102 BUG(); 1102 BUG();
1103 } 1103 }
1104 1104
1105 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth); 1105 tag = find_first_zero_bit(bqt->tag_map, bqt->max_depth);
1106 if (tag >= bqt->max_depth) 1106 if (tag >= bqt->max_depth)
1107 return 1; 1107 return 1;
1108 1108
1109 __set_bit(tag, bqt->tag_map); 1109 __set_bit(tag, bqt->tag_map);
1110 1110
1111 rq->flags |= REQ_QUEUED; 1111 rq->flags |= REQ_QUEUED;
1112 rq->tag = tag; 1112 rq->tag = tag;
1113 bqt->tag_index[tag] = rq; 1113 bqt->tag_index[tag] = rq;
1114 blkdev_dequeue_request(rq); 1114 blkdev_dequeue_request(rq);
1115 list_add(&rq->queuelist, &bqt->busy_list); 1115 list_add(&rq->queuelist, &bqt->busy_list);
1116 bqt->busy++; 1116 bqt->busy++;
1117 return 0; 1117 return 0;
1118 } 1118 }
1119 1119
1120 EXPORT_SYMBOL(blk_queue_start_tag); 1120 EXPORT_SYMBOL(blk_queue_start_tag);
1121 1121
1122 /** 1122 /**
1123 * blk_queue_invalidate_tags - invalidate all pending tags 1123 * blk_queue_invalidate_tags - invalidate all pending tags
1124 * @q: the request queue for the device 1124 * @q: the request queue for the device
1125 * 1125 *
1126 * Description: 1126 * Description:
1127 * Hardware conditions may dictate a need to stop all pending requests. 1127 * Hardware conditions may dictate a need to stop all pending requests.
1128 * In this case, we will safely clear the block side of the tag queue and 1128 * In this case, we will safely clear the block side of the tag queue and
1129 * readd all requests to the request queue in the right order. 1129 * readd all requests to the request queue in the right order.
1130 * 1130 *
1131 * Notes: 1131 * Notes:
1132 * queue lock must be held. 1132 * queue lock must be held.
1133 **/ 1133 **/
1134 void blk_queue_invalidate_tags(request_queue_t *q) 1134 void blk_queue_invalidate_tags(request_queue_t *q)
1135 { 1135 {
1136 struct blk_queue_tag *bqt = q->queue_tags; 1136 struct blk_queue_tag *bqt = q->queue_tags;
1137 struct list_head *tmp, *n; 1137 struct list_head *tmp, *n;
1138 struct request *rq; 1138 struct request *rq;
1139 1139
1140 list_for_each_safe(tmp, n, &bqt->busy_list) { 1140 list_for_each_safe(tmp, n, &bqt->busy_list) {
1141 rq = list_entry_rq(tmp); 1141 rq = list_entry_rq(tmp);
1142 1142
1143 if (rq->tag == -1) { 1143 if (rq->tag == -1) {
1144 printk(KERN_ERR 1144 printk(KERN_ERR
1145 "%s: bad tag found on list\n", __FUNCTION__); 1145 "%s: bad tag found on list\n", __FUNCTION__);
1146 list_del_init(&rq->queuelist); 1146 list_del_init(&rq->queuelist);
1147 rq->flags &= ~REQ_QUEUED; 1147 rq->flags &= ~REQ_QUEUED;
1148 } else 1148 } else
1149 blk_queue_end_tag(q, rq); 1149 blk_queue_end_tag(q, rq);
1150 1150
1151 rq->flags &= ~REQ_STARTED; 1151 rq->flags &= ~REQ_STARTED;
1152 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0); 1152 __elv_add_request(q, rq, ELEVATOR_INSERT_BACK, 0);
1153 } 1153 }
1154 } 1154 }
1155 1155
1156 EXPORT_SYMBOL(blk_queue_invalidate_tags); 1156 EXPORT_SYMBOL(blk_queue_invalidate_tags);
1157 1157
1158 static const char * const rq_flags[] = { 1158 static const char * const rq_flags[] = {
1159 "REQ_RW", 1159 "REQ_RW",
1160 "REQ_FAILFAST", 1160 "REQ_FAILFAST",
1161 "REQ_SORTED", 1161 "REQ_SORTED",
1162 "REQ_SOFTBARRIER", 1162 "REQ_SOFTBARRIER",
1163 "REQ_HARDBARRIER", 1163 "REQ_HARDBARRIER",
1164 "REQ_FUA", 1164 "REQ_FUA",
1165 "REQ_CMD", 1165 "REQ_CMD",
1166 "REQ_NOMERGE", 1166 "REQ_NOMERGE",
1167 "REQ_STARTED", 1167 "REQ_STARTED",
1168 "REQ_DONTPREP", 1168 "REQ_DONTPREP",
1169 "REQ_QUEUED", 1169 "REQ_QUEUED",
1170 "REQ_ELVPRIV", 1170 "REQ_ELVPRIV",
1171 "REQ_PC", 1171 "REQ_PC",
1172 "REQ_BLOCK_PC", 1172 "REQ_BLOCK_PC",
1173 "REQ_SENSE", 1173 "REQ_SENSE",
1174 "REQ_FAILED", 1174 "REQ_FAILED",
1175 "REQ_QUIET", 1175 "REQ_QUIET",
1176 "REQ_SPECIAL", 1176 "REQ_SPECIAL",
1177 "REQ_DRIVE_CMD", 1177 "REQ_DRIVE_CMD",
1178 "REQ_DRIVE_TASK", 1178 "REQ_DRIVE_TASK",
1179 "REQ_DRIVE_TASKFILE", 1179 "REQ_DRIVE_TASKFILE",
1180 "REQ_PREEMPT", 1180 "REQ_PREEMPT",
1181 "REQ_PM_SUSPEND", 1181 "REQ_PM_SUSPEND",
1182 "REQ_PM_RESUME", 1182 "REQ_PM_RESUME",
1183 "REQ_PM_SHUTDOWN", 1183 "REQ_PM_SHUTDOWN",
1184 "REQ_ORDERED_COLOR", 1184 "REQ_ORDERED_COLOR",
1185 }; 1185 };
1186 1186
1187 void blk_dump_rq_flags(struct request *rq, char *msg) 1187 void blk_dump_rq_flags(struct request *rq, char *msg)
1188 { 1188 {
1189 int bit; 1189 int bit;
1190 1190
1191 printk("%s: dev %s: flags = ", msg, 1191 printk("%s: dev %s: flags = ", msg,
1192 rq->rq_disk ? rq->rq_disk->disk_name : "?"); 1192 rq->rq_disk ? rq->rq_disk->disk_name : "?");
1193 bit = 0; 1193 bit = 0;
1194 do { 1194 do {
1195 if (rq->flags & (1 << bit)) 1195 if (rq->flags & (1 << bit))
1196 printk("%s ", rq_flags[bit]); 1196 printk("%s ", rq_flags[bit]);
1197 bit++; 1197 bit++;
1198 } while (bit < __REQ_NR_BITS); 1198 } while (bit < __REQ_NR_BITS);
1199 1199
1200 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector, 1200 printk("\nsector %llu, nr/cnr %lu/%u\n", (unsigned long long)rq->sector,
1201 rq->nr_sectors, 1201 rq->nr_sectors,
1202 rq->current_nr_sectors); 1202 rq->current_nr_sectors);
1203 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len); 1203 printk("bio %p, biotail %p, buffer %p, data %p, len %u\n", rq->bio, rq->biotail, rq->buffer, rq->data, rq->data_len);
1204 1204
1205 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) { 1205 if (rq->flags & (REQ_BLOCK_PC | REQ_PC)) {
1206 printk("cdb: "); 1206 printk("cdb: ");
1207 for (bit = 0; bit < sizeof(rq->cmd); bit++) 1207 for (bit = 0; bit < sizeof(rq->cmd); bit++)
1208 printk("%02x ", rq->cmd[bit]); 1208 printk("%02x ", rq->cmd[bit]);
1209 printk("\n"); 1209 printk("\n");
1210 } 1210 }
1211 } 1211 }
1212 1212
1213 EXPORT_SYMBOL(blk_dump_rq_flags); 1213 EXPORT_SYMBOL(blk_dump_rq_flags);
1214 1214
1215 void blk_recount_segments(request_queue_t *q, struct bio *bio) 1215 void blk_recount_segments(request_queue_t *q, struct bio *bio)
1216 { 1216 {
1217 struct bio_vec *bv, *bvprv = NULL; 1217 struct bio_vec *bv, *bvprv = NULL;
1218 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster; 1218 int i, nr_phys_segs, nr_hw_segs, seg_size, hw_seg_size, cluster;
1219 int high, highprv = 1; 1219 int high, highprv = 1;
1220 1220
1221 if (unlikely(!bio->bi_io_vec)) 1221 if (unlikely(!bio->bi_io_vec))
1222 return; 1222 return;
1223 1223
1224 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); 1224 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1225 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0; 1225 hw_seg_size = seg_size = nr_phys_segs = nr_hw_segs = 0;
1226 bio_for_each_segment(bv, bio, i) { 1226 bio_for_each_segment(bv, bio, i) {
1227 /* 1227 /*
1228 * the trick here is making sure that a high page is never 1228 * the trick here is making sure that a high page is never
1229 * considered part of another segment, since that might 1229 * considered part of another segment, since that might
1230 * change with the bounce page. 1230 * change with the bounce page.
1231 */ 1231 */
1232 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn; 1232 high = page_to_pfn(bv->bv_page) >= q->bounce_pfn;
1233 if (high || highprv) 1233 if (high || highprv)
1234 goto new_hw_segment; 1234 goto new_hw_segment;
1235 if (cluster) { 1235 if (cluster) {
1236 if (seg_size + bv->bv_len > q->max_segment_size) 1236 if (seg_size + bv->bv_len > q->max_segment_size)
1237 goto new_segment; 1237 goto new_segment;
1238 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv)) 1238 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bv))
1239 goto new_segment; 1239 goto new_segment;
1240 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv)) 1240 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bv))
1241 goto new_segment; 1241 goto new_segment;
1242 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) 1242 if (BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len))
1243 goto new_hw_segment; 1243 goto new_hw_segment;
1244 1244
1245 seg_size += bv->bv_len; 1245 seg_size += bv->bv_len;
1246 hw_seg_size += bv->bv_len; 1246 hw_seg_size += bv->bv_len;
1247 bvprv = bv; 1247 bvprv = bv;
1248 continue; 1248 continue;
1249 } 1249 }
1250 new_segment: 1250 new_segment:
1251 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) && 1251 if (BIOVEC_VIRT_MERGEABLE(bvprv, bv) &&
1252 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) { 1252 !BIOVEC_VIRT_OVERSIZE(hw_seg_size + bv->bv_len)) {
1253 hw_seg_size += bv->bv_len; 1253 hw_seg_size += bv->bv_len;
1254 } else { 1254 } else {
1255 new_hw_segment: 1255 new_hw_segment:
1256 if (hw_seg_size > bio->bi_hw_front_size) 1256 if (hw_seg_size > bio->bi_hw_front_size)
1257 bio->bi_hw_front_size = hw_seg_size; 1257 bio->bi_hw_front_size = hw_seg_size;
1258 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len; 1258 hw_seg_size = BIOVEC_VIRT_START_SIZE(bv) + bv->bv_len;
1259 nr_hw_segs++; 1259 nr_hw_segs++;
1260 } 1260 }
1261 1261
1262 nr_phys_segs++; 1262 nr_phys_segs++;
1263 bvprv = bv; 1263 bvprv = bv;
1264 seg_size = bv->bv_len; 1264 seg_size = bv->bv_len;
1265 highprv = high; 1265 highprv = high;
1266 } 1266 }
1267 if (hw_seg_size > bio->bi_hw_back_size) 1267 if (hw_seg_size > bio->bi_hw_back_size)
1268 bio->bi_hw_back_size = hw_seg_size; 1268 bio->bi_hw_back_size = hw_seg_size;
1269 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size) 1269 if (nr_hw_segs == 1 && hw_seg_size > bio->bi_hw_front_size)
1270 bio->bi_hw_front_size = hw_seg_size; 1270 bio->bi_hw_front_size = hw_seg_size;
1271 bio->bi_phys_segments = nr_phys_segs; 1271 bio->bi_phys_segments = nr_phys_segs;
1272 bio->bi_hw_segments = nr_hw_segs; 1272 bio->bi_hw_segments = nr_hw_segs;
1273 bio->bi_flags |= (1 << BIO_SEG_VALID); 1273 bio->bi_flags |= (1 << BIO_SEG_VALID);
1274 } 1274 }
1275 1275
1276 1276
1277 static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio, 1277 static int blk_phys_contig_segment(request_queue_t *q, struct bio *bio,
1278 struct bio *nxt) 1278 struct bio *nxt)
1279 { 1279 {
1280 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER))) 1280 if (!(q->queue_flags & (1 << QUEUE_FLAG_CLUSTER)))
1281 return 0; 1281 return 0;
1282 1282
1283 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt))) 1283 if (!BIOVEC_PHYS_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)))
1284 return 0; 1284 return 0;
1285 if (bio->bi_size + nxt->bi_size > q->max_segment_size) 1285 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1286 return 0; 1286 return 0;
1287 1287
1288 /* 1288 /*
1289 * bio and nxt are contigous in memory, check if the queue allows 1289 * bio and nxt are contigous in memory, check if the queue allows
1290 * these two to be merged into one 1290 * these two to be merged into one
1291 */ 1291 */
1292 if (BIO_SEG_BOUNDARY(q, bio, nxt)) 1292 if (BIO_SEG_BOUNDARY(q, bio, nxt))
1293 return 1; 1293 return 1;
1294 1294
1295 return 0; 1295 return 0;
1296 } 1296 }
1297 1297
1298 static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio, 1298 static int blk_hw_contig_segment(request_queue_t *q, struct bio *bio,
1299 struct bio *nxt) 1299 struct bio *nxt)
1300 { 1300 {
1301 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 1301 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1302 blk_recount_segments(q, bio); 1302 blk_recount_segments(q, bio);
1303 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID))) 1303 if (unlikely(!bio_flagged(nxt, BIO_SEG_VALID)))
1304 blk_recount_segments(q, nxt); 1304 blk_recount_segments(q, nxt);
1305 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) || 1305 if (!BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(nxt)) ||
1306 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size)) 1306 BIOVEC_VIRT_OVERSIZE(bio->bi_hw_front_size + bio->bi_hw_back_size))
1307 return 0; 1307 return 0;
1308 if (bio->bi_size + nxt->bi_size > q->max_segment_size) 1308 if (bio->bi_size + nxt->bi_size > q->max_segment_size)
1309 return 0; 1309 return 0;
1310 1310
1311 return 1; 1311 return 1;
1312 } 1312 }
1313 1313
1314 /* 1314 /*
1315 * map a request to scatterlist, return number of sg entries setup. Caller 1315 * map a request to scatterlist, return number of sg entries setup. Caller
1316 * must make sure sg can hold rq->nr_phys_segments entries 1316 * must make sure sg can hold rq->nr_phys_segments entries
1317 */ 1317 */
1318 int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg) 1318 int blk_rq_map_sg(request_queue_t *q, struct request *rq, struct scatterlist *sg)
1319 { 1319 {
1320 struct bio_vec *bvec, *bvprv; 1320 struct bio_vec *bvec, *bvprv;
1321 struct bio *bio; 1321 struct bio *bio;
1322 int nsegs, i, cluster; 1322 int nsegs, i, cluster;
1323 1323
1324 nsegs = 0; 1324 nsegs = 0;
1325 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER); 1325 cluster = q->queue_flags & (1 << QUEUE_FLAG_CLUSTER);
1326 1326
1327 /* 1327 /*
1328 * for each bio in rq 1328 * for each bio in rq
1329 */ 1329 */
1330 bvprv = NULL; 1330 bvprv = NULL;
1331 rq_for_each_bio(bio, rq) { 1331 rq_for_each_bio(bio, rq) {
1332 /* 1332 /*
1333 * for each segment in bio 1333 * for each segment in bio
1334 */ 1334 */
1335 bio_for_each_segment(bvec, bio, i) { 1335 bio_for_each_segment(bvec, bio, i) {
1336 int nbytes = bvec->bv_len; 1336 int nbytes = bvec->bv_len;
1337 1337
1338 if (bvprv && cluster) { 1338 if (bvprv && cluster) {
1339 if (sg[nsegs - 1].length + nbytes > q->max_segment_size) 1339 if (sg[nsegs - 1].length + nbytes > q->max_segment_size)
1340 goto new_segment; 1340 goto new_segment;
1341 1341
1342 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec)) 1342 if (!BIOVEC_PHYS_MERGEABLE(bvprv, bvec))
1343 goto new_segment; 1343 goto new_segment;
1344 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec)) 1344 if (!BIOVEC_SEG_BOUNDARY(q, bvprv, bvec))
1345 goto new_segment; 1345 goto new_segment;
1346 1346
1347 sg[nsegs - 1].length += nbytes; 1347 sg[nsegs - 1].length += nbytes;
1348 } else { 1348 } else {
1349 new_segment: 1349 new_segment:
1350 memset(&sg[nsegs],0,sizeof(struct scatterlist)); 1350 memset(&sg[nsegs],0,sizeof(struct scatterlist));
1351 sg[nsegs].page = bvec->bv_page; 1351 sg[nsegs].page = bvec->bv_page;
1352 sg[nsegs].length = nbytes; 1352 sg[nsegs].length = nbytes;
1353 sg[nsegs].offset = bvec->bv_offset; 1353 sg[nsegs].offset = bvec->bv_offset;
1354 1354
1355 nsegs++; 1355 nsegs++;
1356 } 1356 }
1357 bvprv = bvec; 1357 bvprv = bvec;
1358 } /* segments in bio */ 1358 } /* segments in bio */
1359 } /* bios in rq */ 1359 } /* bios in rq */
1360 1360
1361 return nsegs; 1361 return nsegs;
1362 } 1362 }
1363 1363
1364 EXPORT_SYMBOL(blk_rq_map_sg); 1364 EXPORT_SYMBOL(blk_rq_map_sg);
1365 1365
1366 /* 1366 /*
1367 * the standard queue merge functions, can be overridden with device 1367 * the standard queue merge functions, can be overridden with device
1368 * specific ones if so desired 1368 * specific ones if so desired
1369 */ 1369 */
1370 1370
1371 static inline int ll_new_mergeable(request_queue_t *q, 1371 static inline int ll_new_mergeable(request_queue_t *q,
1372 struct request *req, 1372 struct request *req,
1373 struct bio *bio) 1373 struct bio *bio)
1374 { 1374 {
1375 int nr_phys_segs = bio_phys_segments(q, bio); 1375 int nr_phys_segs = bio_phys_segments(q, bio);
1376 1376
1377 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { 1377 if (req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1378 req->flags |= REQ_NOMERGE; 1378 req->flags |= REQ_NOMERGE;
1379 if (req == q->last_merge) 1379 if (req == q->last_merge)
1380 q->last_merge = NULL; 1380 q->last_merge = NULL;
1381 return 0; 1381 return 0;
1382 } 1382 }
1383 1383
1384 /* 1384 /*
1385 * A hw segment is just getting larger, bump just the phys 1385 * A hw segment is just getting larger, bump just the phys
1386 * counter. 1386 * counter.
1387 */ 1387 */
1388 req->nr_phys_segments += nr_phys_segs; 1388 req->nr_phys_segments += nr_phys_segs;
1389 return 1; 1389 return 1;
1390 } 1390 }
1391 1391
1392 static inline int ll_new_hw_segment(request_queue_t *q, 1392 static inline int ll_new_hw_segment(request_queue_t *q,
1393 struct request *req, 1393 struct request *req,
1394 struct bio *bio) 1394 struct bio *bio)
1395 { 1395 {
1396 int nr_hw_segs = bio_hw_segments(q, bio); 1396 int nr_hw_segs = bio_hw_segments(q, bio);
1397 int nr_phys_segs = bio_phys_segments(q, bio); 1397 int nr_phys_segs = bio_phys_segments(q, bio);
1398 1398
1399 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments 1399 if (req->nr_hw_segments + nr_hw_segs > q->max_hw_segments
1400 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) { 1400 || req->nr_phys_segments + nr_phys_segs > q->max_phys_segments) {
1401 req->flags |= REQ_NOMERGE; 1401 req->flags |= REQ_NOMERGE;
1402 if (req == q->last_merge) 1402 if (req == q->last_merge)
1403 q->last_merge = NULL; 1403 q->last_merge = NULL;
1404 return 0; 1404 return 0;
1405 } 1405 }
1406 1406
1407 /* 1407 /*
1408 * This will form the start of a new hw segment. Bump both 1408 * This will form the start of a new hw segment. Bump both
1409 * counters. 1409 * counters.
1410 */ 1410 */
1411 req->nr_hw_segments += nr_hw_segs; 1411 req->nr_hw_segments += nr_hw_segs;
1412 req->nr_phys_segments += nr_phys_segs; 1412 req->nr_phys_segments += nr_phys_segs;
1413 return 1; 1413 return 1;
1414 } 1414 }
1415 1415
1416 static int ll_back_merge_fn(request_queue_t *q, struct request *req, 1416 static int ll_back_merge_fn(request_queue_t *q, struct request *req,
1417 struct bio *bio) 1417 struct bio *bio)
1418 { 1418 {
1419 unsigned short max_sectors; 1419 unsigned short max_sectors;
1420 int len; 1420 int len;
1421 1421
1422 if (unlikely(blk_pc_request(req))) 1422 if (unlikely(blk_pc_request(req)))
1423 max_sectors = q->max_hw_sectors; 1423 max_sectors = q->max_hw_sectors;
1424 else 1424 else
1425 max_sectors = q->max_sectors; 1425 max_sectors = q->max_sectors;
1426 1426
1427 if (req->nr_sectors + bio_sectors(bio) > max_sectors) { 1427 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
1428 req->flags |= REQ_NOMERGE; 1428 req->flags |= REQ_NOMERGE;
1429 if (req == q->last_merge) 1429 if (req == q->last_merge)
1430 q->last_merge = NULL; 1430 q->last_merge = NULL;
1431 return 0; 1431 return 0;
1432 } 1432 }
1433 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID))) 1433 if (unlikely(!bio_flagged(req->biotail, BIO_SEG_VALID)))
1434 blk_recount_segments(q, req->biotail); 1434 blk_recount_segments(q, req->biotail);
1435 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 1435 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1436 blk_recount_segments(q, bio); 1436 blk_recount_segments(q, bio);
1437 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size; 1437 len = req->biotail->bi_hw_back_size + bio->bi_hw_front_size;
1438 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) && 1438 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(req->biotail), __BVEC_START(bio)) &&
1439 !BIOVEC_VIRT_OVERSIZE(len)) { 1439 !BIOVEC_VIRT_OVERSIZE(len)) {
1440 int mergeable = ll_new_mergeable(q, req, bio); 1440 int mergeable = ll_new_mergeable(q, req, bio);
1441 1441
1442 if (mergeable) { 1442 if (mergeable) {
1443 if (req->nr_hw_segments == 1) 1443 if (req->nr_hw_segments == 1)
1444 req->bio->bi_hw_front_size = len; 1444 req->bio->bi_hw_front_size = len;
1445 if (bio->bi_hw_segments == 1) 1445 if (bio->bi_hw_segments == 1)
1446 bio->bi_hw_back_size = len; 1446 bio->bi_hw_back_size = len;
1447 } 1447 }
1448 return mergeable; 1448 return mergeable;
1449 } 1449 }
1450 1450
1451 return ll_new_hw_segment(q, req, bio); 1451 return ll_new_hw_segment(q, req, bio);
1452 } 1452 }
1453 1453
1454 static int ll_front_merge_fn(request_queue_t *q, struct request *req, 1454 static int ll_front_merge_fn(request_queue_t *q, struct request *req,
1455 struct bio *bio) 1455 struct bio *bio)
1456 { 1456 {
1457 unsigned short max_sectors; 1457 unsigned short max_sectors;
1458 int len; 1458 int len;
1459 1459
1460 if (unlikely(blk_pc_request(req))) 1460 if (unlikely(blk_pc_request(req)))
1461 max_sectors = q->max_hw_sectors; 1461 max_sectors = q->max_hw_sectors;
1462 else 1462 else
1463 max_sectors = q->max_sectors; 1463 max_sectors = q->max_sectors;
1464 1464
1465 1465
1466 if (req->nr_sectors + bio_sectors(bio) > max_sectors) { 1466 if (req->nr_sectors + bio_sectors(bio) > max_sectors) {
1467 req->flags |= REQ_NOMERGE; 1467 req->flags |= REQ_NOMERGE;
1468 if (req == q->last_merge) 1468 if (req == q->last_merge)
1469 q->last_merge = NULL; 1469 q->last_merge = NULL;
1470 return 0; 1470 return 0;
1471 } 1471 }
1472 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size; 1472 len = bio->bi_hw_back_size + req->bio->bi_hw_front_size;
1473 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 1473 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
1474 blk_recount_segments(q, bio); 1474 blk_recount_segments(q, bio);
1475 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID))) 1475 if (unlikely(!bio_flagged(req->bio, BIO_SEG_VALID)))
1476 blk_recount_segments(q, req->bio); 1476 blk_recount_segments(q, req->bio);
1477 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) && 1477 if (BIOVEC_VIRT_MERGEABLE(__BVEC_END(bio), __BVEC_START(req->bio)) &&
1478 !BIOVEC_VIRT_OVERSIZE(len)) { 1478 !BIOVEC_VIRT_OVERSIZE(len)) {
1479 int mergeable = ll_new_mergeable(q, req, bio); 1479 int mergeable = ll_new_mergeable(q, req, bio);
1480 1480
1481 if (mergeable) { 1481 if (mergeable) {
1482 if (bio->bi_hw_segments == 1) 1482 if (bio->bi_hw_segments == 1)
1483 bio->bi_hw_front_size = len; 1483 bio->bi_hw_front_size = len;
1484 if (req->nr_hw_segments == 1) 1484 if (req->nr_hw_segments == 1)
1485 req->biotail->bi_hw_back_size = len; 1485 req->biotail->bi_hw_back_size = len;
1486 } 1486 }
1487 return mergeable; 1487 return mergeable;
1488 } 1488 }
1489 1489
1490 return ll_new_hw_segment(q, req, bio); 1490 return ll_new_hw_segment(q, req, bio);
1491 } 1491 }
1492 1492
1493 static int ll_merge_requests_fn(request_queue_t *q, struct request *req, 1493 static int ll_merge_requests_fn(request_queue_t *q, struct request *req,
1494 struct request *next) 1494 struct request *next)
1495 { 1495 {
1496 int total_phys_segments; 1496 int total_phys_segments;
1497 int total_hw_segments; 1497 int total_hw_segments;
1498 1498
1499 /* 1499 /*
1500 * First check if the either of the requests are re-queued 1500 * First check if the either of the requests are re-queued
1501 * requests. Can't merge them if they are. 1501 * requests. Can't merge them if they are.
1502 */ 1502 */
1503 if (req->special || next->special) 1503 if (req->special || next->special)
1504 return 0; 1504 return 0;
1505 1505
1506 /* 1506 /*
1507 * Will it become too large? 1507 * Will it become too large?
1508 */ 1508 */
1509 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors) 1509 if ((req->nr_sectors + next->nr_sectors) > q->max_sectors)
1510 return 0; 1510 return 0;
1511 1511
1512 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments; 1512 total_phys_segments = req->nr_phys_segments + next->nr_phys_segments;
1513 if (blk_phys_contig_segment(q, req->biotail, next->bio)) 1513 if (blk_phys_contig_segment(q, req->biotail, next->bio))
1514 total_phys_segments--; 1514 total_phys_segments--;
1515 1515
1516 if (total_phys_segments > q->max_phys_segments) 1516 if (total_phys_segments > q->max_phys_segments)
1517 return 0; 1517 return 0;
1518 1518
1519 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments; 1519 total_hw_segments = req->nr_hw_segments + next->nr_hw_segments;
1520 if (blk_hw_contig_segment(q, req->biotail, next->bio)) { 1520 if (blk_hw_contig_segment(q, req->biotail, next->bio)) {
1521 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size; 1521 int len = req->biotail->bi_hw_back_size + next->bio->bi_hw_front_size;
1522 /* 1522 /*
1523 * propagate the combined length to the end of the requests 1523 * propagate the combined length to the end of the requests
1524 */ 1524 */
1525 if (req->nr_hw_segments == 1) 1525 if (req->nr_hw_segments == 1)
1526 req->bio->bi_hw_front_size = len; 1526 req->bio->bi_hw_front_size = len;
1527 if (next->nr_hw_segments == 1) 1527 if (next->nr_hw_segments == 1)
1528 next->biotail->bi_hw_back_size = len; 1528 next->biotail->bi_hw_back_size = len;
1529 total_hw_segments--; 1529 total_hw_segments--;
1530 } 1530 }
1531 1531
1532 if (total_hw_segments > q->max_hw_segments) 1532 if (total_hw_segments > q->max_hw_segments)
1533 return 0; 1533 return 0;
1534 1534
1535 /* Merge is OK... */ 1535 /* Merge is OK... */
1536 req->nr_phys_segments = total_phys_segments; 1536 req->nr_phys_segments = total_phys_segments;
1537 req->nr_hw_segments = total_hw_segments; 1537 req->nr_hw_segments = total_hw_segments;
1538 return 1; 1538 return 1;
1539 } 1539 }
1540 1540
1541 /* 1541 /*
1542 * "plug" the device if there are no outstanding requests: this will 1542 * "plug" the device if there are no outstanding requests: this will
1543 * force the transfer to start only after we have put all the requests 1543 * force the transfer to start only after we have put all the requests
1544 * on the list. 1544 * on the list.
1545 * 1545 *
1546 * This is called with interrupts off and no requests on the queue and 1546 * This is called with interrupts off and no requests on the queue and
1547 * with the queue lock held. 1547 * with the queue lock held.
1548 */ 1548 */
1549 void blk_plug_device(request_queue_t *q) 1549 void blk_plug_device(request_queue_t *q)
1550 { 1550 {
1551 WARN_ON(!irqs_disabled()); 1551 WARN_ON(!irqs_disabled());
1552 1552
1553 /* 1553 /*
1554 * don't plug a stopped queue, it must be paired with blk_start_queue() 1554 * don't plug a stopped queue, it must be paired with blk_start_queue()
1555 * which will restart the queueing 1555 * which will restart the queueing
1556 */ 1556 */
1557 if (blk_queue_stopped(q)) 1557 if (blk_queue_stopped(q))
1558 return; 1558 return;
1559 1559
1560 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) { 1560 if (!test_and_set_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) {
1561 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay); 1561 mod_timer(&q->unplug_timer, jiffies + q->unplug_delay);
1562 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG); 1562 blk_add_trace_generic(q, NULL, 0, BLK_TA_PLUG);
1563 } 1563 }
1564 } 1564 }
1565 1565
1566 EXPORT_SYMBOL(blk_plug_device); 1566 EXPORT_SYMBOL(blk_plug_device);
1567 1567
1568 /* 1568 /*
1569 * remove the queue from the plugged list, if present. called with 1569 * remove the queue from the plugged list, if present. called with
1570 * queue lock held and interrupts disabled. 1570 * queue lock held and interrupts disabled.
1571 */ 1571 */
1572 int blk_remove_plug(request_queue_t *q) 1572 int blk_remove_plug(request_queue_t *q)
1573 { 1573 {
1574 WARN_ON(!irqs_disabled()); 1574 WARN_ON(!irqs_disabled());
1575 1575
1576 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags)) 1576 if (!test_and_clear_bit(QUEUE_FLAG_PLUGGED, &q->queue_flags))
1577 return 0; 1577 return 0;
1578 1578
1579 del_timer(&q->unplug_timer); 1579 del_timer(&q->unplug_timer);
1580 return 1; 1580 return 1;
1581 } 1581 }
1582 1582
1583 EXPORT_SYMBOL(blk_remove_plug); 1583 EXPORT_SYMBOL(blk_remove_plug);
1584 1584
1585 /* 1585 /*
1586 * remove the plug and let it rip.. 1586 * remove the plug and let it rip..
1587 */ 1587 */
1588 void __generic_unplug_device(request_queue_t *q) 1588 void __generic_unplug_device(request_queue_t *q)
1589 { 1589 {
1590 if (unlikely(blk_queue_stopped(q))) 1590 if (unlikely(blk_queue_stopped(q)))
1591 return; 1591 return;
1592 1592
1593 if (!blk_remove_plug(q)) 1593 if (!blk_remove_plug(q))
1594 return; 1594 return;
1595 1595
1596 q->request_fn(q); 1596 q->request_fn(q);
1597 } 1597 }
1598 EXPORT_SYMBOL(__generic_unplug_device); 1598 EXPORT_SYMBOL(__generic_unplug_device);
1599 1599
1600 /** 1600 /**
1601 * generic_unplug_device - fire a request queue 1601 * generic_unplug_device - fire a request queue
1602 * @q: The &request_queue_t in question 1602 * @q: The &request_queue_t in question
1603 * 1603 *
1604 * Description: 1604 * Description:
1605 * Linux uses plugging to build bigger requests queues before letting 1605 * Linux uses plugging to build bigger requests queues before letting
1606 * the device have at them. If a queue is plugged, the I/O scheduler 1606 * the device have at them. If a queue is plugged, the I/O scheduler
1607 * is still adding and merging requests on the queue. Once the queue 1607 * is still adding and merging requests on the queue. Once the queue
1608 * gets unplugged, the request_fn defined for the queue is invoked and 1608 * gets unplugged, the request_fn defined for the queue is invoked and
1609 * transfers started. 1609 * transfers started.
1610 **/ 1610 **/
1611 void generic_unplug_device(request_queue_t *q) 1611 void generic_unplug_device(request_queue_t *q)
1612 { 1612 {
1613 spin_lock_irq(q->queue_lock); 1613 spin_lock_irq(q->queue_lock);
1614 __generic_unplug_device(q); 1614 __generic_unplug_device(q);
1615 spin_unlock_irq(q->queue_lock); 1615 spin_unlock_irq(q->queue_lock);
1616 } 1616 }
1617 EXPORT_SYMBOL(generic_unplug_device); 1617 EXPORT_SYMBOL(generic_unplug_device);
1618 1618
1619 static void blk_backing_dev_unplug(struct backing_dev_info *bdi, 1619 static void blk_backing_dev_unplug(struct backing_dev_info *bdi,
1620 struct page *page) 1620 struct page *page)
1621 { 1621 {
1622 request_queue_t *q = bdi->unplug_io_data; 1622 request_queue_t *q = bdi->unplug_io_data;
1623 1623
1624 /* 1624 /*
1625 * devices don't necessarily have an ->unplug_fn defined 1625 * devices don't necessarily have an ->unplug_fn defined
1626 */ 1626 */
1627 if (q->unplug_fn) { 1627 if (q->unplug_fn) {
1628 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, 1628 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1629 q->rq.count[READ] + q->rq.count[WRITE]); 1629 q->rq.count[READ] + q->rq.count[WRITE]);
1630 1630
1631 q->unplug_fn(q); 1631 q->unplug_fn(q);
1632 } 1632 }
1633 } 1633 }
1634 1634
1635 static void blk_unplug_work(void *data) 1635 static void blk_unplug_work(void *data)
1636 { 1636 {
1637 request_queue_t *q = data; 1637 request_queue_t *q = data;
1638 1638
1639 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL, 1639 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_IO, NULL,
1640 q->rq.count[READ] + q->rq.count[WRITE]); 1640 q->rq.count[READ] + q->rq.count[WRITE]);
1641 1641
1642 q->unplug_fn(q); 1642 q->unplug_fn(q);
1643 } 1643 }
1644 1644
1645 static void blk_unplug_timeout(unsigned long data) 1645 static void blk_unplug_timeout(unsigned long data)
1646 { 1646 {
1647 request_queue_t *q = (request_queue_t *)data; 1647 request_queue_t *q = (request_queue_t *)data;
1648 1648
1649 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL, 1649 blk_add_trace_pdu_int(q, BLK_TA_UNPLUG_TIMER, NULL,
1650 q->rq.count[READ] + q->rq.count[WRITE]); 1650 q->rq.count[READ] + q->rq.count[WRITE]);
1651 1651
1652 kblockd_schedule_work(&q->unplug_work); 1652 kblockd_schedule_work(&q->unplug_work);
1653 } 1653 }
1654 1654
1655 /** 1655 /**
1656 * blk_start_queue - restart a previously stopped queue 1656 * blk_start_queue - restart a previously stopped queue
1657 * @q: The &request_queue_t in question 1657 * @q: The &request_queue_t in question
1658 * 1658 *
1659 * Description: 1659 * Description:
1660 * blk_start_queue() will clear the stop flag on the queue, and call 1660 * blk_start_queue() will clear the stop flag on the queue, and call
1661 * the request_fn for the queue if it was in a stopped state when 1661 * the request_fn for the queue if it was in a stopped state when
1662 * entered. Also see blk_stop_queue(). Queue lock must be held. 1662 * entered. Also see blk_stop_queue(). Queue lock must be held.
1663 **/ 1663 **/
1664 void blk_start_queue(request_queue_t *q) 1664 void blk_start_queue(request_queue_t *q)
1665 { 1665 {
1666 WARN_ON(!irqs_disabled());
1667
1666 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); 1668 clear_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1667 1669
1668 /* 1670 /*
1669 * one level of recursion is ok and is much faster than kicking 1671 * one level of recursion is ok and is much faster than kicking
1670 * the unplug handling 1672 * the unplug handling
1671 */ 1673 */
1672 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { 1674 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1673 q->request_fn(q); 1675 q->request_fn(q);
1674 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); 1676 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1675 } else { 1677 } else {
1676 blk_plug_device(q); 1678 blk_plug_device(q);
1677 kblockd_schedule_work(&q->unplug_work); 1679 kblockd_schedule_work(&q->unplug_work);
1678 } 1680 }
1679 } 1681 }
1680 1682
1681 EXPORT_SYMBOL(blk_start_queue); 1683 EXPORT_SYMBOL(blk_start_queue);
1682 1684
1683 /** 1685 /**
1684 * blk_stop_queue - stop a queue 1686 * blk_stop_queue - stop a queue
1685 * @q: The &request_queue_t in question 1687 * @q: The &request_queue_t in question
1686 * 1688 *
1687 * Description: 1689 * Description:
1688 * The Linux block layer assumes that a block driver will consume all 1690 * The Linux block layer assumes that a block driver will consume all
1689 * entries on the request queue when the request_fn strategy is called. 1691 * entries on the request queue when the request_fn strategy is called.
1690 * Often this will not happen, because of hardware limitations (queue 1692 * Often this will not happen, because of hardware limitations (queue
1691 * depth settings). If a device driver gets a 'queue full' response, 1693 * depth settings). If a device driver gets a 'queue full' response,
1692 * or if it simply chooses not to queue more I/O at one point, it can 1694 * or if it simply chooses not to queue more I/O at one point, it can
1693 * call this function to prevent the request_fn from being called until 1695 * call this function to prevent the request_fn from being called until
1694 * the driver has signalled it's ready to go again. This happens by calling 1696 * the driver has signalled it's ready to go again. This happens by calling
1695 * blk_start_queue() to restart queue operations. Queue lock must be held. 1697 * blk_start_queue() to restart queue operations. Queue lock must be held.
1696 **/ 1698 **/
1697 void blk_stop_queue(request_queue_t *q) 1699 void blk_stop_queue(request_queue_t *q)
1698 { 1700 {
1699 blk_remove_plug(q); 1701 blk_remove_plug(q);
1700 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags); 1702 set_bit(QUEUE_FLAG_STOPPED, &q->queue_flags);
1701 } 1703 }
1702 EXPORT_SYMBOL(blk_stop_queue); 1704 EXPORT_SYMBOL(blk_stop_queue);
1703 1705
1704 /** 1706 /**
1705 * blk_sync_queue - cancel any pending callbacks on a queue 1707 * blk_sync_queue - cancel any pending callbacks on a queue
1706 * @q: the queue 1708 * @q: the queue
1707 * 1709 *
1708 * Description: 1710 * Description:
1709 * The block layer may perform asynchronous callback activity 1711 * The block layer may perform asynchronous callback activity
1710 * on a queue, such as calling the unplug function after a timeout. 1712 * on a queue, such as calling the unplug function after a timeout.
1711 * A block device may call blk_sync_queue to ensure that any 1713 * A block device may call blk_sync_queue to ensure that any
1712 * such activity is cancelled, thus allowing it to release resources 1714 * such activity is cancelled, thus allowing it to release resources
1713 * the the callbacks might use. The caller must already have made sure 1715 * the the callbacks might use. The caller must already have made sure
1714 * that its ->make_request_fn will not re-add plugging prior to calling 1716 * that its ->make_request_fn will not re-add plugging prior to calling
1715 * this function. 1717 * this function.
1716 * 1718 *
1717 */ 1719 */
1718 void blk_sync_queue(struct request_queue *q) 1720 void blk_sync_queue(struct request_queue *q)
1719 { 1721 {
1720 del_timer_sync(&q->unplug_timer); 1722 del_timer_sync(&q->unplug_timer);
1721 kblockd_flush(); 1723 kblockd_flush();
1722 } 1724 }
1723 EXPORT_SYMBOL(blk_sync_queue); 1725 EXPORT_SYMBOL(blk_sync_queue);
1724 1726
1725 /** 1727 /**
1726 * blk_run_queue - run a single device queue 1728 * blk_run_queue - run a single device queue
1727 * @q: The queue to run 1729 * @q: The queue to run
1728 */ 1730 */
1729 void blk_run_queue(struct request_queue *q) 1731 void blk_run_queue(struct request_queue *q)
1730 { 1732 {
1731 unsigned long flags; 1733 unsigned long flags;
1732 1734
1733 spin_lock_irqsave(q->queue_lock, flags); 1735 spin_lock_irqsave(q->queue_lock, flags);
1734 blk_remove_plug(q); 1736 blk_remove_plug(q);
1735 1737
1736 /* 1738 /*
1737 * Only recurse once to avoid overrunning the stack, let the unplug 1739 * Only recurse once to avoid overrunning the stack, let the unplug
1738 * handling reinvoke the handler shortly if we already got there. 1740 * handling reinvoke the handler shortly if we already got there.
1739 */ 1741 */
1740 if (!elv_queue_empty(q)) { 1742 if (!elv_queue_empty(q)) {
1741 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) { 1743 if (!test_and_set_bit(QUEUE_FLAG_REENTER, &q->queue_flags)) {
1742 q->request_fn(q); 1744 q->request_fn(q);
1743 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags); 1745 clear_bit(QUEUE_FLAG_REENTER, &q->queue_flags);
1744 } else { 1746 } else {
1745 blk_plug_device(q); 1747 blk_plug_device(q);
1746 kblockd_schedule_work(&q->unplug_work); 1748 kblockd_schedule_work(&q->unplug_work);
1747 } 1749 }
1748 } 1750 }
1749 1751
1750 spin_unlock_irqrestore(q->queue_lock, flags); 1752 spin_unlock_irqrestore(q->queue_lock, flags);
1751 } 1753 }
1752 EXPORT_SYMBOL(blk_run_queue); 1754 EXPORT_SYMBOL(blk_run_queue);
1753 1755
1754 /** 1756 /**
1755 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed 1757 * blk_cleanup_queue: - release a &request_queue_t when it is no longer needed
1756 * @kobj: the kobj belonging of the request queue to be released 1758 * @kobj: the kobj belonging of the request queue to be released
1757 * 1759 *
1758 * Description: 1760 * Description:
1759 * blk_cleanup_queue is the pair to blk_init_queue() or 1761 * blk_cleanup_queue is the pair to blk_init_queue() or
1760 * blk_queue_make_request(). It should be called when a request queue is 1762 * blk_queue_make_request(). It should be called when a request queue is
1761 * being released; typically when a block device is being de-registered. 1763 * being released; typically when a block device is being de-registered.
1762 * Currently, its primary task it to free all the &struct request 1764 * Currently, its primary task it to free all the &struct request
1763 * structures that were allocated to the queue and the queue itself. 1765 * structures that were allocated to the queue and the queue itself.
1764 * 1766 *
1765 * Caveat: 1767 * Caveat:
1766 * Hopefully the low level driver will have finished any 1768 * Hopefully the low level driver will have finished any
1767 * outstanding requests first... 1769 * outstanding requests first...
1768 **/ 1770 **/
1769 static void blk_release_queue(struct kobject *kobj) 1771 static void blk_release_queue(struct kobject *kobj)
1770 { 1772 {
1771 request_queue_t *q = container_of(kobj, struct request_queue, kobj); 1773 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
1772 struct request_list *rl = &q->rq; 1774 struct request_list *rl = &q->rq;
1773 1775
1774 blk_sync_queue(q); 1776 blk_sync_queue(q);
1775 1777
1776 if (rl->rq_pool) 1778 if (rl->rq_pool)
1777 mempool_destroy(rl->rq_pool); 1779 mempool_destroy(rl->rq_pool);
1778 1780
1779 if (q->queue_tags) 1781 if (q->queue_tags)
1780 __blk_queue_free_tags(q); 1782 __blk_queue_free_tags(q);
1781 1783
1782 if (q->blk_trace) 1784 if (q->blk_trace)
1783 blk_trace_shutdown(q); 1785 blk_trace_shutdown(q);
1784 1786
1785 kmem_cache_free(requestq_cachep, q); 1787 kmem_cache_free(requestq_cachep, q);
1786 } 1788 }
1787 1789
1788 void blk_put_queue(request_queue_t *q) 1790 void blk_put_queue(request_queue_t *q)
1789 { 1791 {
1790 kobject_put(&q->kobj); 1792 kobject_put(&q->kobj);
1791 } 1793 }
1792 EXPORT_SYMBOL(blk_put_queue); 1794 EXPORT_SYMBOL(blk_put_queue);
1793 1795
1794 void blk_cleanup_queue(request_queue_t * q) 1796 void blk_cleanup_queue(request_queue_t * q)
1795 { 1797 {
1796 mutex_lock(&q->sysfs_lock); 1798 mutex_lock(&q->sysfs_lock);
1797 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags); 1799 set_bit(QUEUE_FLAG_DEAD, &q->queue_flags);
1798 mutex_unlock(&q->sysfs_lock); 1800 mutex_unlock(&q->sysfs_lock);
1799 1801
1800 if (q->elevator) 1802 if (q->elevator)
1801 elevator_exit(q->elevator); 1803 elevator_exit(q->elevator);
1802 1804
1803 blk_put_queue(q); 1805 blk_put_queue(q);
1804 } 1806 }
1805 1807
1806 EXPORT_SYMBOL(blk_cleanup_queue); 1808 EXPORT_SYMBOL(blk_cleanup_queue);
1807 1809
1808 static int blk_init_free_list(request_queue_t *q) 1810 static int blk_init_free_list(request_queue_t *q)
1809 { 1811 {
1810 struct request_list *rl = &q->rq; 1812 struct request_list *rl = &q->rq;
1811 1813
1812 rl->count[READ] = rl->count[WRITE] = 0; 1814 rl->count[READ] = rl->count[WRITE] = 0;
1813 rl->starved[READ] = rl->starved[WRITE] = 0; 1815 rl->starved[READ] = rl->starved[WRITE] = 0;
1814 rl->elvpriv = 0; 1816 rl->elvpriv = 0;
1815 init_waitqueue_head(&rl->wait[READ]); 1817 init_waitqueue_head(&rl->wait[READ]);
1816 init_waitqueue_head(&rl->wait[WRITE]); 1818 init_waitqueue_head(&rl->wait[WRITE]);
1817 1819
1818 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, 1820 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
1819 mempool_free_slab, request_cachep, q->node); 1821 mempool_free_slab, request_cachep, q->node);
1820 1822
1821 if (!rl->rq_pool) 1823 if (!rl->rq_pool)
1822 return -ENOMEM; 1824 return -ENOMEM;
1823 1825
1824 return 0; 1826 return 0;
1825 } 1827 }
1826 1828
1827 request_queue_t *blk_alloc_queue(gfp_t gfp_mask) 1829 request_queue_t *blk_alloc_queue(gfp_t gfp_mask)
1828 { 1830 {
1829 return blk_alloc_queue_node(gfp_mask, -1); 1831 return blk_alloc_queue_node(gfp_mask, -1);
1830 } 1832 }
1831 EXPORT_SYMBOL(blk_alloc_queue); 1833 EXPORT_SYMBOL(blk_alloc_queue);
1832 1834
1833 static struct kobj_type queue_ktype; 1835 static struct kobj_type queue_ktype;
1834 1836
1835 request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) 1837 request_queue_t *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
1836 { 1838 {
1837 request_queue_t *q; 1839 request_queue_t *q;
1838 1840
1839 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id); 1841 q = kmem_cache_alloc_node(requestq_cachep, gfp_mask, node_id);
1840 if (!q) 1842 if (!q)
1841 return NULL; 1843 return NULL;
1842 1844
1843 memset(q, 0, sizeof(*q)); 1845 memset(q, 0, sizeof(*q));
1844 init_timer(&q->unplug_timer); 1846 init_timer(&q->unplug_timer);
1845 1847
1846 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue"); 1848 snprintf(q->kobj.name, KOBJ_NAME_LEN, "%s", "queue");
1847 q->kobj.ktype = &queue_ktype; 1849 q->kobj.ktype = &queue_ktype;
1848 kobject_init(&q->kobj); 1850 kobject_init(&q->kobj);
1849 1851
1850 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug; 1852 q->backing_dev_info.unplug_io_fn = blk_backing_dev_unplug;
1851 q->backing_dev_info.unplug_io_data = q; 1853 q->backing_dev_info.unplug_io_data = q;
1852 1854
1853 mutex_init(&q->sysfs_lock); 1855 mutex_init(&q->sysfs_lock);
1854 1856
1855 return q; 1857 return q;
1856 } 1858 }
1857 EXPORT_SYMBOL(blk_alloc_queue_node); 1859 EXPORT_SYMBOL(blk_alloc_queue_node);
1858 1860
1859 /** 1861 /**
1860 * blk_init_queue - prepare a request queue for use with a block device 1862 * blk_init_queue - prepare a request queue for use with a block device
1861 * @rfn: The function to be called to process requests that have been 1863 * @rfn: The function to be called to process requests that have been
1862 * placed on the queue. 1864 * placed on the queue.
1863 * @lock: Request queue spin lock 1865 * @lock: Request queue spin lock
1864 * 1866 *
1865 * Description: 1867 * Description:
1866 * If a block device wishes to use the standard request handling procedures, 1868 * If a block device wishes to use the standard request handling procedures,
1867 * which sorts requests and coalesces adjacent requests, then it must 1869 * which sorts requests and coalesces adjacent requests, then it must
1868 * call blk_init_queue(). The function @rfn will be called when there 1870 * call blk_init_queue(). The function @rfn will be called when there
1869 * are requests on the queue that need to be processed. If the device 1871 * are requests on the queue that need to be processed. If the device
1870 * supports plugging, then @rfn may not be called immediately when requests 1872 * supports plugging, then @rfn may not be called immediately when requests
1871 * are available on the queue, but may be called at some time later instead. 1873 * are available on the queue, but may be called at some time later instead.
1872 * Plugged queues are generally unplugged when a buffer belonging to one 1874 * Plugged queues are generally unplugged when a buffer belonging to one
1873 * of the requests on the queue is needed, or due to memory pressure. 1875 * of the requests on the queue is needed, or due to memory pressure.
1874 * 1876 *
1875 * @rfn is not required, or even expected, to remove all requests off the 1877 * @rfn is not required, or even expected, to remove all requests off the
1876 * queue, but only as many as it can handle at a time. If it does leave 1878 * queue, but only as many as it can handle at a time. If it does leave
1877 * requests on the queue, it is responsible for arranging that the requests 1879 * requests on the queue, it is responsible for arranging that the requests
1878 * get dealt with eventually. 1880 * get dealt with eventually.
1879 * 1881 *
1880 * The queue spin lock must be held while manipulating the requests on the 1882 * The queue spin lock must be held while manipulating the requests on the
1881 * request queue. 1883 * request queue; this lock will be taken also from interrupt context, so irq
1884 * disabling is needed for it.
1882 * 1885 *
1883 * Function returns a pointer to the initialized request queue, or NULL if 1886 * Function returns a pointer to the initialized request queue, or NULL if
1884 * it didn't succeed. 1887 * it didn't succeed.
1885 * 1888 *
1886 * Note: 1889 * Note:
1887 * blk_init_queue() must be paired with a blk_cleanup_queue() call 1890 * blk_init_queue() must be paired with a blk_cleanup_queue() call
1888 * when the block device is deactivated (such as at module unload). 1891 * when the block device is deactivated (such as at module unload).
1889 **/ 1892 **/
1890 1893
1891 request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) 1894 request_queue_t *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
1892 { 1895 {
1893 return blk_init_queue_node(rfn, lock, -1); 1896 return blk_init_queue_node(rfn, lock, -1);
1894 } 1897 }
1895 EXPORT_SYMBOL(blk_init_queue); 1898 EXPORT_SYMBOL(blk_init_queue);
1896 1899
1897 request_queue_t * 1900 request_queue_t *
1898 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) 1901 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
1899 { 1902 {
1900 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id); 1903 request_queue_t *q = blk_alloc_queue_node(GFP_KERNEL, node_id);
1901 1904
1902 if (!q) 1905 if (!q)
1903 return NULL; 1906 return NULL;
1904 1907
1905 q->node = node_id; 1908 q->node = node_id;
1906 if (blk_init_free_list(q)) { 1909 if (blk_init_free_list(q)) {
1907 kmem_cache_free(requestq_cachep, q); 1910 kmem_cache_free(requestq_cachep, q);
1908 return NULL; 1911 return NULL;
1909 } 1912 }
1910 1913
1911 /* 1914 /*
1912 * if caller didn't supply a lock, they get per-queue locking with 1915 * if caller didn't supply a lock, they get per-queue locking with
1913 * our embedded lock 1916 * our embedded lock
1914 */ 1917 */
1915 if (!lock) { 1918 if (!lock) {
1916 spin_lock_init(&q->__queue_lock); 1919 spin_lock_init(&q->__queue_lock);
1917 lock = &q->__queue_lock; 1920 lock = &q->__queue_lock;
1918 } 1921 }
1919 1922
1920 q->request_fn = rfn; 1923 q->request_fn = rfn;
1921 q->back_merge_fn = ll_back_merge_fn; 1924 q->back_merge_fn = ll_back_merge_fn;
1922 q->front_merge_fn = ll_front_merge_fn; 1925 q->front_merge_fn = ll_front_merge_fn;
1923 q->merge_requests_fn = ll_merge_requests_fn; 1926 q->merge_requests_fn = ll_merge_requests_fn;
1924 q->prep_rq_fn = NULL; 1927 q->prep_rq_fn = NULL;
1925 q->unplug_fn = generic_unplug_device; 1928 q->unplug_fn = generic_unplug_device;
1926 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER); 1929 q->queue_flags = (1 << QUEUE_FLAG_CLUSTER);
1927 q->queue_lock = lock; 1930 q->queue_lock = lock;
1928 1931
1929 blk_queue_segment_boundary(q, 0xffffffff); 1932 blk_queue_segment_boundary(q, 0xffffffff);
1930 1933
1931 blk_queue_make_request(q, __make_request); 1934 blk_queue_make_request(q, __make_request);
1932 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE); 1935 blk_queue_max_segment_size(q, MAX_SEGMENT_SIZE);
1933 1936
1934 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS); 1937 blk_queue_max_hw_segments(q, MAX_HW_SEGMENTS);
1935 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS); 1938 blk_queue_max_phys_segments(q, MAX_PHYS_SEGMENTS);
1936 1939
1937 /* 1940 /*
1938 * all done 1941 * all done
1939 */ 1942 */
1940 if (!elevator_init(q, NULL)) { 1943 if (!elevator_init(q, NULL)) {
1941 blk_queue_congestion_threshold(q); 1944 blk_queue_congestion_threshold(q);
1942 return q; 1945 return q;
1943 } 1946 }
1944 1947
1945 blk_put_queue(q); 1948 blk_put_queue(q);
1946 return NULL; 1949 return NULL;
1947 } 1950 }
1948 EXPORT_SYMBOL(blk_init_queue_node); 1951 EXPORT_SYMBOL(blk_init_queue_node);
1949 1952
1950 int blk_get_queue(request_queue_t *q) 1953 int blk_get_queue(request_queue_t *q)
1951 { 1954 {
1952 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) { 1955 if (likely(!test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) {
1953 kobject_get(&q->kobj); 1956 kobject_get(&q->kobj);
1954 return 0; 1957 return 0;
1955 } 1958 }
1956 1959
1957 return 1; 1960 return 1;
1958 } 1961 }
1959 1962
1960 EXPORT_SYMBOL(blk_get_queue); 1963 EXPORT_SYMBOL(blk_get_queue);
1961 1964
1962 static inline void blk_free_request(request_queue_t *q, struct request *rq) 1965 static inline void blk_free_request(request_queue_t *q, struct request *rq)
1963 { 1966 {
1964 if (rq->flags & REQ_ELVPRIV) 1967 if (rq->flags & REQ_ELVPRIV)
1965 elv_put_request(q, rq); 1968 elv_put_request(q, rq);
1966 mempool_free(rq, q->rq.rq_pool); 1969 mempool_free(rq, q->rq.rq_pool);
1967 } 1970 }
1968 1971
1969 static inline struct request * 1972 static inline struct request *
1970 blk_alloc_request(request_queue_t *q, int rw, struct bio *bio, 1973 blk_alloc_request(request_queue_t *q, int rw, struct bio *bio,
1971 int priv, gfp_t gfp_mask) 1974 int priv, gfp_t gfp_mask)
1972 { 1975 {
1973 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask); 1976 struct request *rq = mempool_alloc(q->rq.rq_pool, gfp_mask);
1974 1977
1975 if (!rq) 1978 if (!rq)
1976 return NULL; 1979 return NULL;
1977 1980
1978 /* 1981 /*
1979 * first three bits are identical in rq->flags and bio->bi_rw, 1982 * first three bits are identical in rq->flags and bio->bi_rw,
1980 * see bio.h and blkdev.h 1983 * see bio.h and blkdev.h
1981 */ 1984 */
1982 rq->flags = rw; 1985 rq->flags = rw;
1983 1986
1984 if (priv) { 1987 if (priv) {
1985 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) { 1988 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) {
1986 mempool_free(rq, q->rq.rq_pool); 1989 mempool_free(rq, q->rq.rq_pool);
1987 return NULL; 1990 return NULL;
1988 } 1991 }
1989 rq->flags |= REQ_ELVPRIV; 1992 rq->flags |= REQ_ELVPRIV;
1990 } 1993 }
1991 1994
1992 return rq; 1995 return rq;
1993 } 1996 }
1994 1997
1995 /* 1998 /*
1996 * ioc_batching returns true if the ioc is a valid batching request and 1999 * ioc_batching returns true if the ioc is a valid batching request and
1997 * should be given priority access to a request. 2000 * should be given priority access to a request.
1998 */ 2001 */
1999 static inline int ioc_batching(request_queue_t *q, struct io_context *ioc) 2002 static inline int ioc_batching(request_queue_t *q, struct io_context *ioc)
2000 { 2003 {
2001 if (!ioc) 2004 if (!ioc)
2002 return 0; 2005 return 0;
2003 2006
2004 /* 2007 /*
2005 * Make sure the process is able to allocate at least 1 request 2008 * Make sure the process is able to allocate at least 1 request
2006 * even if the batch times out, otherwise we could theoretically 2009 * even if the batch times out, otherwise we could theoretically
2007 * lose wakeups. 2010 * lose wakeups.
2008 */ 2011 */
2009 return ioc->nr_batch_requests == q->nr_batching || 2012 return ioc->nr_batch_requests == q->nr_batching ||
2010 (ioc->nr_batch_requests > 0 2013 (ioc->nr_batch_requests > 0
2011 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); 2014 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
2012 } 2015 }
2013 2016
2014 /* 2017 /*
2015 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This 2018 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
2016 * will cause the process to be a "batcher" on all queues in the system. This 2019 * will cause the process to be a "batcher" on all queues in the system. This
2017 * is the behaviour we want though - once it gets a wakeup it should be given 2020 * is the behaviour we want though - once it gets a wakeup it should be given
2018 * a nice run. 2021 * a nice run.
2019 */ 2022 */
2020 static void ioc_set_batching(request_queue_t *q, struct io_context *ioc) 2023 static void ioc_set_batching(request_queue_t *q, struct io_context *ioc)
2021 { 2024 {
2022 if (!ioc || ioc_batching(q, ioc)) 2025 if (!ioc || ioc_batching(q, ioc))
2023 return; 2026 return;
2024 2027
2025 ioc->nr_batch_requests = q->nr_batching; 2028 ioc->nr_batch_requests = q->nr_batching;
2026 ioc->last_waited = jiffies; 2029 ioc->last_waited = jiffies;
2027 } 2030 }
2028 2031
2029 static void __freed_request(request_queue_t *q, int rw) 2032 static void __freed_request(request_queue_t *q, int rw)
2030 { 2033 {
2031 struct request_list *rl = &q->rq; 2034 struct request_list *rl = &q->rq;
2032 2035
2033 if (rl->count[rw] < queue_congestion_off_threshold(q)) 2036 if (rl->count[rw] < queue_congestion_off_threshold(q))
2034 clear_queue_congested(q, rw); 2037 clear_queue_congested(q, rw);
2035 2038
2036 if (rl->count[rw] + 1 <= q->nr_requests) { 2039 if (rl->count[rw] + 1 <= q->nr_requests) {
2037 if (waitqueue_active(&rl->wait[rw])) 2040 if (waitqueue_active(&rl->wait[rw]))
2038 wake_up(&rl->wait[rw]); 2041 wake_up(&rl->wait[rw]);
2039 2042
2040 blk_clear_queue_full(q, rw); 2043 blk_clear_queue_full(q, rw);
2041 } 2044 }
2042 } 2045 }
2043 2046
2044 /* 2047 /*
2045 * A request has just been released. Account for it, update the full and 2048 * A request has just been released. Account for it, update the full and
2046 * congestion status, wake up any waiters. Called under q->queue_lock. 2049 * congestion status, wake up any waiters. Called under q->queue_lock.
2047 */ 2050 */
2048 static void freed_request(request_queue_t *q, int rw, int priv) 2051 static void freed_request(request_queue_t *q, int rw, int priv)
2049 { 2052 {
2050 struct request_list *rl = &q->rq; 2053 struct request_list *rl = &q->rq;
2051 2054
2052 rl->count[rw]--; 2055 rl->count[rw]--;
2053 if (priv) 2056 if (priv)
2054 rl->elvpriv--; 2057 rl->elvpriv--;
2055 2058
2056 __freed_request(q, rw); 2059 __freed_request(q, rw);
2057 2060
2058 if (unlikely(rl->starved[rw ^ 1])) 2061 if (unlikely(rl->starved[rw ^ 1]))
2059 __freed_request(q, rw ^ 1); 2062 __freed_request(q, rw ^ 1);
2060 } 2063 }
2061 2064
2062 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist) 2065 #define blkdev_free_rq(list) list_entry((list)->next, struct request, queuelist)
2063 /* 2066 /*
2064 * Get a free request, queue_lock must be held. 2067 * Get a free request, queue_lock must be held.
2065 * Returns NULL on failure, with queue_lock held. 2068 * Returns NULL on failure, with queue_lock held.
2066 * Returns !NULL on success, with queue_lock *not held*. 2069 * Returns !NULL on success, with queue_lock *not held*.
2067 */ 2070 */
2068 static struct request *get_request(request_queue_t *q, int rw, struct bio *bio, 2071 static struct request *get_request(request_queue_t *q, int rw, struct bio *bio,
2069 gfp_t gfp_mask) 2072 gfp_t gfp_mask)
2070 { 2073 {
2071 struct request *rq = NULL; 2074 struct request *rq = NULL;
2072 struct request_list *rl = &q->rq; 2075 struct request_list *rl = &q->rq;
2073 struct io_context *ioc = NULL; 2076 struct io_context *ioc = NULL;
2074 int may_queue, priv; 2077 int may_queue, priv;
2075 2078
2076 may_queue = elv_may_queue(q, rw, bio); 2079 may_queue = elv_may_queue(q, rw, bio);
2077 if (may_queue == ELV_MQUEUE_NO) 2080 if (may_queue == ELV_MQUEUE_NO)
2078 goto rq_starved; 2081 goto rq_starved;
2079 2082
2080 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) { 2083 if (rl->count[rw]+1 >= queue_congestion_on_threshold(q)) {
2081 if (rl->count[rw]+1 >= q->nr_requests) { 2084 if (rl->count[rw]+1 >= q->nr_requests) {
2082 ioc = current_io_context(GFP_ATOMIC); 2085 ioc = current_io_context(GFP_ATOMIC);
2083 /* 2086 /*
2084 * The queue will fill after this allocation, so set 2087 * The queue will fill after this allocation, so set
2085 * it as full, and mark this process as "batching". 2088 * it as full, and mark this process as "batching".
2086 * This process will be allowed to complete a batch of 2089 * This process will be allowed to complete a batch of
2087 * requests, others will be blocked. 2090 * requests, others will be blocked.
2088 */ 2091 */
2089 if (!blk_queue_full(q, rw)) { 2092 if (!blk_queue_full(q, rw)) {
2090 ioc_set_batching(q, ioc); 2093 ioc_set_batching(q, ioc);
2091 blk_set_queue_full(q, rw); 2094 blk_set_queue_full(q, rw);
2092 } else { 2095 } else {
2093 if (may_queue != ELV_MQUEUE_MUST 2096 if (may_queue != ELV_MQUEUE_MUST
2094 && !ioc_batching(q, ioc)) { 2097 && !ioc_batching(q, ioc)) {
2095 /* 2098 /*
2096 * The queue is full and the allocating 2099 * The queue is full and the allocating
2097 * process is not a "batcher", and not 2100 * process is not a "batcher", and not
2098 * exempted by the IO scheduler 2101 * exempted by the IO scheduler
2099 */ 2102 */
2100 goto out; 2103 goto out;
2101 } 2104 }
2102 } 2105 }
2103 } 2106 }
2104 set_queue_congested(q, rw); 2107 set_queue_congested(q, rw);
2105 } 2108 }
2106 2109
2107 /* 2110 /*
2108 * Only allow batching queuers to allocate up to 50% over the defined 2111 * Only allow batching queuers to allocate up to 50% over the defined
2109 * limit of requests, otherwise we could have thousands of requests 2112 * limit of requests, otherwise we could have thousands of requests
2110 * allocated with any setting of ->nr_requests 2113 * allocated with any setting of ->nr_requests
2111 */ 2114 */
2112 if (rl->count[rw] >= (3 * q->nr_requests / 2)) 2115 if (rl->count[rw] >= (3 * q->nr_requests / 2))
2113 goto out; 2116 goto out;
2114 2117
2115 rl->count[rw]++; 2118 rl->count[rw]++;
2116 rl->starved[rw] = 0; 2119 rl->starved[rw] = 0;
2117 2120
2118 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); 2121 priv = !test_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
2119 if (priv) 2122 if (priv)
2120 rl->elvpriv++; 2123 rl->elvpriv++;
2121 2124
2122 spin_unlock_irq(q->queue_lock); 2125 spin_unlock_irq(q->queue_lock);
2123 2126
2124 rq = blk_alloc_request(q, rw, bio, priv, gfp_mask); 2127 rq = blk_alloc_request(q, rw, bio, priv, gfp_mask);
2125 if (unlikely(!rq)) { 2128 if (unlikely(!rq)) {
2126 /* 2129 /*
2127 * Allocation failed presumably due to memory. Undo anything 2130 * Allocation failed presumably due to memory. Undo anything
2128 * we might have messed up. 2131 * we might have messed up.
2129 * 2132 *
2130 * Allocating task should really be put onto the front of the 2133 * Allocating task should really be put onto the front of the
2131 * wait queue, but this is pretty rare. 2134 * wait queue, but this is pretty rare.
2132 */ 2135 */
2133 spin_lock_irq(q->queue_lock); 2136 spin_lock_irq(q->queue_lock);
2134 freed_request(q, rw, priv); 2137 freed_request(q, rw, priv);
2135 2138
2136 /* 2139 /*
2137 * in the very unlikely event that allocation failed and no 2140 * in the very unlikely event that allocation failed and no
2138 * requests for this direction was pending, mark us starved 2141 * requests for this direction was pending, mark us starved
2139 * so that freeing of a request in the other direction will 2142 * so that freeing of a request in the other direction will
2140 * notice us. another possible fix would be to split the 2143 * notice us. another possible fix would be to split the
2141 * rq mempool into READ and WRITE 2144 * rq mempool into READ and WRITE
2142 */ 2145 */
2143 rq_starved: 2146 rq_starved:
2144 if (unlikely(rl->count[rw] == 0)) 2147 if (unlikely(rl->count[rw] == 0))
2145 rl->starved[rw] = 1; 2148 rl->starved[rw] = 1;
2146 2149
2147 goto out; 2150 goto out;
2148 } 2151 }
2149 2152
2150 /* 2153 /*
2151 * ioc may be NULL here, and ioc_batching will be false. That's 2154 * ioc may be NULL here, and ioc_batching will be false. That's
2152 * OK, if the queue is under the request limit then requests need 2155 * OK, if the queue is under the request limit then requests need
2153 * not count toward the nr_batch_requests limit. There will always 2156 * not count toward the nr_batch_requests limit. There will always
2154 * be some limit enforced by BLK_BATCH_TIME. 2157 * be some limit enforced by BLK_BATCH_TIME.
2155 */ 2158 */
2156 if (ioc_batching(q, ioc)) 2159 if (ioc_batching(q, ioc))
2157 ioc->nr_batch_requests--; 2160 ioc->nr_batch_requests--;
2158 2161
2159 rq_init(q, rq); 2162 rq_init(q, rq);
2160 rq->rl = rl; 2163 rq->rl = rl;
2161 2164
2162 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ); 2165 blk_add_trace_generic(q, bio, rw, BLK_TA_GETRQ);
2163 out: 2166 out:
2164 return rq; 2167 return rq;
2165 } 2168 }
2166 2169
2167 /* 2170 /*
2168 * No available requests for this queue, unplug the device and wait for some 2171 * No available requests for this queue, unplug the device and wait for some
2169 * requests to become available. 2172 * requests to become available.
2170 * 2173 *
2171 * Called with q->queue_lock held, and returns with it unlocked. 2174 * Called with q->queue_lock held, and returns with it unlocked.
2172 */ 2175 */
2173 static struct request *get_request_wait(request_queue_t *q, int rw, 2176 static struct request *get_request_wait(request_queue_t *q, int rw,
2174 struct bio *bio) 2177 struct bio *bio)
2175 { 2178 {
2176 struct request *rq; 2179 struct request *rq;
2177 2180
2178 rq = get_request(q, rw, bio, GFP_NOIO); 2181 rq = get_request(q, rw, bio, GFP_NOIO);
2179 while (!rq) { 2182 while (!rq) {
2180 DEFINE_WAIT(wait); 2183 DEFINE_WAIT(wait);
2181 struct request_list *rl = &q->rq; 2184 struct request_list *rl = &q->rq;
2182 2185
2183 prepare_to_wait_exclusive(&rl->wait[rw], &wait, 2186 prepare_to_wait_exclusive(&rl->wait[rw], &wait,
2184 TASK_UNINTERRUPTIBLE); 2187 TASK_UNINTERRUPTIBLE);
2185 2188
2186 rq = get_request(q, rw, bio, GFP_NOIO); 2189 rq = get_request(q, rw, bio, GFP_NOIO);
2187 2190
2188 if (!rq) { 2191 if (!rq) {
2189 struct io_context *ioc; 2192 struct io_context *ioc;
2190 2193
2191 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ); 2194 blk_add_trace_generic(q, bio, rw, BLK_TA_SLEEPRQ);
2192 2195
2193 __generic_unplug_device(q); 2196 __generic_unplug_device(q);
2194 spin_unlock_irq(q->queue_lock); 2197 spin_unlock_irq(q->queue_lock);
2195 io_schedule(); 2198 io_schedule();
2196 2199
2197 /* 2200 /*
2198 * After sleeping, we become a "batching" process and 2201 * After sleeping, we become a "batching" process and
2199 * will be able to allocate at least one request, and 2202 * will be able to allocate at least one request, and
2200 * up to a big batch of them for a small period time. 2203 * up to a big batch of them for a small period time.
2201 * See ioc_batching, ioc_set_batching 2204 * See ioc_batching, ioc_set_batching
2202 */ 2205 */
2203 ioc = current_io_context(GFP_NOIO); 2206 ioc = current_io_context(GFP_NOIO);
2204 ioc_set_batching(q, ioc); 2207 ioc_set_batching(q, ioc);
2205 2208
2206 spin_lock_irq(q->queue_lock); 2209 spin_lock_irq(q->queue_lock);
2207 } 2210 }
2208 finish_wait(&rl->wait[rw], &wait); 2211 finish_wait(&rl->wait[rw], &wait);
2209 } 2212 }
2210 2213
2211 return rq; 2214 return rq;
2212 } 2215 }
2213 2216
2214 struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask) 2217 struct request *blk_get_request(request_queue_t *q, int rw, gfp_t gfp_mask)
2215 { 2218 {
2216 struct request *rq; 2219 struct request *rq;
2217 2220
2218 BUG_ON(rw != READ && rw != WRITE); 2221 BUG_ON(rw != READ && rw != WRITE);
2219 2222
2220 spin_lock_irq(q->queue_lock); 2223 spin_lock_irq(q->queue_lock);
2221 if (gfp_mask & __GFP_WAIT) { 2224 if (gfp_mask & __GFP_WAIT) {
2222 rq = get_request_wait(q, rw, NULL); 2225 rq = get_request_wait(q, rw, NULL);
2223 } else { 2226 } else {
2224 rq = get_request(q, rw, NULL, gfp_mask); 2227 rq = get_request(q, rw, NULL, gfp_mask);
2225 if (!rq) 2228 if (!rq)
2226 spin_unlock_irq(q->queue_lock); 2229 spin_unlock_irq(q->queue_lock);
2227 } 2230 }
2228 /* q->queue_lock is unlocked at this point */ 2231 /* q->queue_lock is unlocked at this point */
2229 2232
2230 return rq; 2233 return rq;
2231 } 2234 }
2232 EXPORT_SYMBOL(blk_get_request); 2235 EXPORT_SYMBOL(blk_get_request);
2233 2236
2234 /** 2237 /**
2235 * blk_requeue_request - put a request back on queue 2238 * blk_requeue_request - put a request back on queue
2236 * @q: request queue where request should be inserted 2239 * @q: request queue where request should be inserted
2237 * @rq: request to be inserted 2240 * @rq: request to be inserted
2238 * 2241 *
2239 * Description: 2242 * Description:
2240 * Drivers often keep queueing requests until the hardware cannot accept 2243 * Drivers often keep queueing requests until the hardware cannot accept
2241 * more, when that condition happens we need to put the request back 2244 * more, when that condition happens we need to put the request back
2242 * on the queue. Must be called with queue lock held. 2245 * on the queue. Must be called with queue lock held.
2243 */ 2246 */
2244 void blk_requeue_request(request_queue_t *q, struct request *rq) 2247 void blk_requeue_request(request_queue_t *q, struct request *rq)
2245 { 2248 {
2246 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); 2249 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
2247 2250
2248 if (blk_rq_tagged(rq)) 2251 if (blk_rq_tagged(rq))
2249 blk_queue_end_tag(q, rq); 2252 blk_queue_end_tag(q, rq);
2250 2253
2251 elv_requeue_request(q, rq); 2254 elv_requeue_request(q, rq);
2252 } 2255 }
2253 2256
2254 EXPORT_SYMBOL(blk_requeue_request); 2257 EXPORT_SYMBOL(blk_requeue_request);
2255 2258
2256 /** 2259 /**
2257 * blk_insert_request - insert a special request in to a request queue 2260 * blk_insert_request - insert a special request in to a request queue
2258 * @q: request queue where request should be inserted 2261 * @q: request queue where request should be inserted
2259 * @rq: request to be inserted 2262 * @rq: request to be inserted
2260 * @at_head: insert request at head or tail of queue 2263 * @at_head: insert request at head or tail of queue
2261 * @data: private data 2264 * @data: private data
2262 * 2265 *
2263 * Description: 2266 * Description:
2264 * Many block devices need to execute commands asynchronously, so they don't 2267 * Many block devices need to execute commands asynchronously, so they don't
2265 * block the whole kernel from preemption during request execution. This is 2268 * block the whole kernel from preemption during request execution. This is
2266 * accomplished normally by inserting aritficial requests tagged as 2269 * accomplished normally by inserting aritficial requests tagged as
2267 * REQ_SPECIAL in to the corresponding request queue, and letting them be 2270 * REQ_SPECIAL in to the corresponding request queue, and letting them be
2268 * scheduled for actual execution by the request queue. 2271 * scheduled for actual execution by the request queue.
2269 * 2272 *
2270 * We have the option of inserting the head or the tail of the queue. 2273 * We have the option of inserting the head or the tail of the queue.
2271 * Typically we use the tail for new ioctls and so forth. We use the head 2274 * Typically we use the tail for new ioctls and so forth. We use the head
2272 * of the queue for things like a QUEUE_FULL message from a device, or a 2275 * of the queue for things like a QUEUE_FULL message from a device, or a
2273 * host that is unable to accept a particular command. 2276 * host that is unable to accept a particular command.
2274 */ 2277 */
2275 void blk_insert_request(request_queue_t *q, struct request *rq, 2278 void blk_insert_request(request_queue_t *q, struct request *rq,
2276 int at_head, void *data) 2279 int at_head, void *data)
2277 { 2280 {
2278 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; 2281 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2279 unsigned long flags; 2282 unsigned long flags;
2280 2283
2281 /* 2284 /*
2282 * tell I/O scheduler that this isn't a regular read/write (ie it 2285 * tell I/O scheduler that this isn't a regular read/write (ie it
2283 * must not attempt merges on this) and that it acts as a soft 2286 * must not attempt merges on this) and that it acts as a soft
2284 * barrier 2287 * barrier
2285 */ 2288 */
2286 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER; 2289 rq->flags |= REQ_SPECIAL | REQ_SOFTBARRIER;
2287 2290
2288 rq->special = data; 2291 rq->special = data;
2289 2292
2290 spin_lock_irqsave(q->queue_lock, flags); 2293 spin_lock_irqsave(q->queue_lock, flags);
2291 2294
2292 /* 2295 /*
2293 * If command is tagged, release the tag 2296 * If command is tagged, release the tag
2294 */ 2297 */
2295 if (blk_rq_tagged(rq)) 2298 if (blk_rq_tagged(rq))
2296 blk_queue_end_tag(q, rq); 2299 blk_queue_end_tag(q, rq);
2297 2300
2298 drive_stat_acct(rq, rq->nr_sectors, 1); 2301 drive_stat_acct(rq, rq->nr_sectors, 1);
2299 __elv_add_request(q, rq, where, 0); 2302 __elv_add_request(q, rq, where, 0);
2300 2303
2301 if (blk_queue_plugged(q)) 2304 if (blk_queue_plugged(q))
2302 __generic_unplug_device(q); 2305 __generic_unplug_device(q);
2303 else 2306 else
2304 q->request_fn(q); 2307 q->request_fn(q);
2305 spin_unlock_irqrestore(q->queue_lock, flags); 2308 spin_unlock_irqrestore(q->queue_lock, flags);
2306 } 2309 }
2307 2310
2308 EXPORT_SYMBOL(blk_insert_request); 2311 EXPORT_SYMBOL(blk_insert_request);
2309 2312
2310 /** 2313 /**
2311 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage 2314 * blk_rq_map_user - map user data to a request, for REQ_BLOCK_PC usage
2312 * @q: request queue where request should be inserted 2315 * @q: request queue where request should be inserted
2313 * @rq: request structure to fill 2316 * @rq: request structure to fill
2314 * @ubuf: the user buffer 2317 * @ubuf: the user buffer
2315 * @len: length of user data 2318 * @len: length of user data
2316 * 2319 *
2317 * Description: 2320 * Description:
2318 * Data will be mapped directly for zero copy io, if possible. Otherwise 2321 * Data will be mapped directly for zero copy io, if possible. Otherwise
2319 * a kernel bounce buffer is used. 2322 * a kernel bounce buffer is used.
2320 * 2323 *
2321 * A matching blk_rq_unmap_user() must be issued at the end of io, while 2324 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2322 * still in process context. 2325 * still in process context.
2323 * 2326 *
2324 * Note: The mapped bio may need to be bounced through blk_queue_bounce() 2327 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2325 * before being submitted to the device, as pages mapped may be out of 2328 * before being submitted to the device, as pages mapped may be out of
2326 * reach. It's the callers responsibility to make sure this happens. The 2329 * reach. It's the callers responsibility to make sure this happens. The
2327 * original bio must be passed back in to blk_rq_unmap_user() for proper 2330 * original bio must be passed back in to blk_rq_unmap_user() for proper
2328 * unmapping. 2331 * unmapping.
2329 */ 2332 */
2330 int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf, 2333 int blk_rq_map_user(request_queue_t *q, struct request *rq, void __user *ubuf,
2331 unsigned int len) 2334 unsigned int len)
2332 { 2335 {
2333 unsigned long uaddr; 2336 unsigned long uaddr;
2334 struct bio *bio; 2337 struct bio *bio;
2335 int reading; 2338 int reading;
2336 2339
2337 if (len > (q->max_hw_sectors << 9)) 2340 if (len > (q->max_hw_sectors << 9))
2338 return -EINVAL; 2341 return -EINVAL;
2339 if (!len || !ubuf) 2342 if (!len || !ubuf)
2340 return -EINVAL; 2343 return -EINVAL;
2341 2344
2342 reading = rq_data_dir(rq) == READ; 2345 reading = rq_data_dir(rq) == READ;
2343 2346
2344 /* 2347 /*
2345 * if alignment requirement is satisfied, map in user pages for 2348 * if alignment requirement is satisfied, map in user pages for
2346 * direct dma. else, set up kernel bounce buffers 2349 * direct dma. else, set up kernel bounce buffers
2347 */ 2350 */
2348 uaddr = (unsigned long) ubuf; 2351 uaddr = (unsigned long) ubuf;
2349 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q))) 2352 if (!(uaddr & queue_dma_alignment(q)) && !(len & queue_dma_alignment(q)))
2350 bio = bio_map_user(q, NULL, uaddr, len, reading); 2353 bio = bio_map_user(q, NULL, uaddr, len, reading);
2351 else 2354 else
2352 bio = bio_copy_user(q, uaddr, len, reading); 2355 bio = bio_copy_user(q, uaddr, len, reading);
2353 2356
2354 if (!IS_ERR(bio)) { 2357 if (!IS_ERR(bio)) {
2355 rq->bio = rq->biotail = bio; 2358 rq->bio = rq->biotail = bio;
2356 blk_rq_bio_prep(q, rq, bio); 2359 blk_rq_bio_prep(q, rq, bio);
2357 2360
2358 rq->buffer = rq->data = NULL; 2361 rq->buffer = rq->data = NULL;
2359 rq->data_len = len; 2362 rq->data_len = len;
2360 return 0; 2363 return 0;
2361 } 2364 }
2362 2365
2363 /* 2366 /*
2364 * bio is the err-ptr 2367 * bio is the err-ptr
2365 */ 2368 */
2366 return PTR_ERR(bio); 2369 return PTR_ERR(bio);
2367 } 2370 }
2368 2371
2369 EXPORT_SYMBOL(blk_rq_map_user); 2372 EXPORT_SYMBOL(blk_rq_map_user);
2370 2373
2371 /** 2374 /**
2372 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage 2375 * blk_rq_map_user_iov - map user data to a request, for REQ_BLOCK_PC usage
2373 * @q: request queue where request should be inserted 2376 * @q: request queue where request should be inserted
2374 * @rq: request to map data to 2377 * @rq: request to map data to
2375 * @iov: pointer to the iovec 2378 * @iov: pointer to the iovec
2376 * @iov_count: number of elements in the iovec 2379 * @iov_count: number of elements in the iovec
2377 * 2380 *
2378 * Description: 2381 * Description:
2379 * Data will be mapped directly for zero copy io, if possible. Otherwise 2382 * Data will be mapped directly for zero copy io, if possible. Otherwise
2380 * a kernel bounce buffer is used. 2383 * a kernel bounce buffer is used.
2381 * 2384 *
2382 * A matching blk_rq_unmap_user() must be issued at the end of io, while 2385 * A matching blk_rq_unmap_user() must be issued at the end of io, while
2383 * still in process context. 2386 * still in process context.
2384 * 2387 *
2385 * Note: The mapped bio may need to be bounced through blk_queue_bounce() 2388 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
2386 * before being submitted to the device, as pages mapped may be out of 2389 * before being submitted to the device, as pages mapped may be out of
2387 * reach. It's the callers responsibility to make sure this happens. The 2390 * reach. It's the callers responsibility to make sure this happens. The
2388 * original bio must be passed back in to blk_rq_unmap_user() for proper 2391 * original bio must be passed back in to blk_rq_unmap_user() for proper
2389 * unmapping. 2392 * unmapping.
2390 */ 2393 */
2391 int blk_rq_map_user_iov(request_queue_t *q, struct request *rq, 2394 int blk_rq_map_user_iov(request_queue_t *q, struct request *rq,
2392 struct sg_iovec *iov, int iov_count) 2395 struct sg_iovec *iov, int iov_count)
2393 { 2396 {
2394 struct bio *bio; 2397 struct bio *bio;
2395 2398
2396 if (!iov || iov_count <= 0) 2399 if (!iov || iov_count <= 0)
2397 return -EINVAL; 2400 return -EINVAL;
2398 2401
2399 /* we don't allow misaligned data like bio_map_user() does. If the 2402 /* we don't allow misaligned data like bio_map_user() does. If the
2400 * user is using sg, they're expected to know the alignment constraints 2403 * user is using sg, they're expected to know the alignment constraints
2401 * and respect them accordingly */ 2404 * and respect them accordingly */
2402 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ); 2405 bio = bio_map_user_iov(q, NULL, iov, iov_count, rq_data_dir(rq)== READ);
2403 if (IS_ERR(bio)) 2406 if (IS_ERR(bio))
2404 return PTR_ERR(bio); 2407 return PTR_ERR(bio);
2405 2408
2406 rq->bio = rq->biotail = bio; 2409 rq->bio = rq->biotail = bio;
2407 blk_rq_bio_prep(q, rq, bio); 2410 blk_rq_bio_prep(q, rq, bio);
2408 rq->buffer = rq->data = NULL; 2411 rq->buffer = rq->data = NULL;
2409 rq->data_len = bio->bi_size; 2412 rq->data_len = bio->bi_size;
2410 return 0; 2413 return 0;
2411 } 2414 }
2412 2415
2413 EXPORT_SYMBOL(blk_rq_map_user_iov); 2416 EXPORT_SYMBOL(blk_rq_map_user_iov);
2414 2417
2415 /** 2418 /**
2416 * blk_rq_unmap_user - unmap a request with user data 2419 * blk_rq_unmap_user - unmap a request with user data
2417 * @bio: bio to be unmapped 2420 * @bio: bio to be unmapped
2418 * @ulen: length of user buffer 2421 * @ulen: length of user buffer
2419 * 2422 *
2420 * Description: 2423 * Description:
2421 * Unmap a bio previously mapped by blk_rq_map_user(). 2424 * Unmap a bio previously mapped by blk_rq_map_user().
2422 */ 2425 */
2423 int blk_rq_unmap_user(struct bio *bio, unsigned int ulen) 2426 int blk_rq_unmap_user(struct bio *bio, unsigned int ulen)
2424 { 2427 {
2425 int ret = 0; 2428 int ret = 0;
2426 2429
2427 if (bio) { 2430 if (bio) {
2428 if (bio_flagged(bio, BIO_USER_MAPPED)) 2431 if (bio_flagged(bio, BIO_USER_MAPPED))
2429 bio_unmap_user(bio); 2432 bio_unmap_user(bio);
2430 else 2433 else
2431 ret = bio_uncopy_user(bio); 2434 ret = bio_uncopy_user(bio);
2432 } 2435 }
2433 2436
2434 return 0; 2437 return 0;
2435 } 2438 }
2436 2439
2437 EXPORT_SYMBOL(blk_rq_unmap_user); 2440 EXPORT_SYMBOL(blk_rq_unmap_user);
2438 2441
2439 /** 2442 /**
2440 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage 2443 * blk_rq_map_kern - map kernel data to a request, for REQ_BLOCK_PC usage
2441 * @q: request queue where request should be inserted 2444 * @q: request queue where request should be inserted
2442 * @rq: request to fill 2445 * @rq: request to fill
2443 * @kbuf: the kernel buffer 2446 * @kbuf: the kernel buffer
2444 * @len: length of user data 2447 * @len: length of user data
2445 * @gfp_mask: memory allocation flags 2448 * @gfp_mask: memory allocation flags
2446 */ 2449 */
2447 int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf, 2450 int blk_rq_map_kern(request_queue_t *q, struct request *rq, void *kbuf,
2448 unsigned int len, gfp_t gfp_mask) 2451 unsigned int len, gfp_t gfp_mask)
2449 { 2452 {
2450 struct bio *bio; 2453 struct bio *bio;
2451 2454
2452 if (len > (q->max_hw_sectors << 9)) 2455 if (len > (q->max_hw_sectors << 9))
2453 return -EINVAL; 2456 return -EINVAL;
2454 if (!len || !kbuf) 2457 if (!len || !kbuf)
2455 return -EINVAL; 2458 return -EINVAL;
2456 2459
2457 bio = bio_map_kern(q, kbuf, len, gfp_mask); 2460 bio = bio_map_kern(q, kbuf, len, gfp_mask);
2458 if (IS_ERR(bio)) 2461 if (IS_ERR(bio))
2459 return PTR_ERR(bio); 2462 return PTR_ERR(bio);
2460 2463
2461 if (rq_data_dir(rq) == WRITE) 2464 if (rq_data_dir(rq) == WRITE)
2462 bio->bi_rw |= (1 << BIO_RW); 2465 bio->bi_rw |= (1 << BIO_RW);
2463 2466
2464 rq->bio = rq->biotail = bio; 2467 rq->bio = rq->biotail = bio;
2465 blk_rq_bio_prep(q, rq, bio); 2468 blk_rq_bio_prep(q, rq, bio);
2466 2469
2467 rq->buffer = rq->data = NULL; 2470 rq->buffer = rq->data = NULL;
2468 rq->data_len = len; 2471 rq->data_len = len;
2469 return 0; 2472 return 0;
2470 } 2473 }
2471 2474
2472 EXPORT_SYMBOL(blk_rq_map_kern); 2475 EXPORT_SYMBOL(blk_rq_map_kern);
2473 2476
2474 /** 2477 /**
2475 * blk_execute_rq_nowait - insert a request into queue for execution 2478 * blk_execute_rq_nowait - insert a request into queue for execution
2476 * @q: queue to insert the request in 2479 * @q: queue to insert the request in
2477 * @bd_disk: matching gendisk 2480 * @bd_disk: matching gendisk
2478 * @rq: request to insert 2481 * @rq: request to insert
2479 * @at_head: insert request at head or tail of queue 2482 * @at_head: insert request at head or tail of queue
2480 * @done: I/O completion handler 2483 * @done: I/O completion handler
2481 * 2484 *
2482 * Description: 2485 * Description:
2483 * Insert a fully prepared request at the back of the io scheduler queue 2486 * Insert a fully prepared request at the back of the io scheduler queue
2484 * for execution. Don't wait for completion. 2487 * for execution. Don't wait for completion.
2485 */ 2488 */
2486 void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk, 2489 void blk_execute_rq_nowait(request_queue_t *q, struct gendisk *bd_disk,
2487 struct request *rq, int at_head, 2490 struct request *rq, int at_head,
2488 rq_end_io_fn *done) 2491 rq_end_io_fn *done)
2489 { 2492 {
2490 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK; 2493 int where = at_head ? ELEVATOR_INSERT_FRONT : ELEVATOR_INSERT_BACK;
2491 2494
2492 rq->rq_disk = bd_disk; 2495 rq->rq_disk = bd_disk;
2493 rq->flags |= REQ_NOMERGE; 2496 rq->flags |= REQ_NOMERGE;
2494 rq->end_io = done; 2497 rq->end_io = done;
2495 WARN_ON(irqs_disabled()); 2498 WARN_ON(irqs_disabled());
2496 spin_lock_irq(q->queue_lock); 2499 spin_lock_irq(q->queue_lock);
2497 __elv_add_request(q, rq, where, 1); 2500 __elv_add_request(q, rq, where, 1);
2498 __generic_unplug_device(q); 2501 __generic_unplug_device(q);
2499 spin_unlock_irq(q->queue_lock); 2502 spin_unlock_irq(q->queue_lock);
2500 } 2503 }
2501 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait); 2504 EXPORT_SYMBOL_GPL(blk_execute_rq_nowait);
2502 2505
2503 /** 2506 /**
2504 * blk_execute_rq - insert a request into queue for execution 2507 * blk_execute_rq - insert a request into queue for execution
2505 * @q: queue to insert the request in 2508 * @q: queue to insert the request in
2506 * @bd_disk: matching gendisk 2509 * @bd_disk: matching gendisk
2507 * @rq: request to insert 2510 * @rq: request to insert
2508 * @at_head: insert request at head or tail of queue 2511 * @at_head: insert request at head or tail of queue
2509 * 2512 *
2510 * Description: 2513 * Description:
2511 * Insert a fully prepared request at the back of the io scheduler queue 2514 * Insert a fully prepared request at the back of the io scheduler queue
2512 * for execution and wait for completion. 2515 * for execution and wait for completion.
2513 */ 2516 */
2514 int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk, 2517 int blk_execute_rq(request_queue_t *q, struct gendisk *bd_disk,
2515 struct request *rq, int at_head) 2518 struct request *rq, int at_head)
2516 { 2519 {
2517 DECLARE_COMPLETION(wait); 2520 DECLARE_COMPLETION(wait);
2518 char sense[SCSI_SENSE_BUFFERSIZE]; 2521 char sense[SCSI_SENSE_BUFFERSIZE];
2519 int err = 0; 2522 int err = 0;
2520 2523
2521 /* 2524 /*
2522 * we need an extra reference to the request, so we can look at 2525 * we need an extra reference to the request, so we can look at
2523 * it after io completion 2526 * it after io completion
2524 */ 2527 */
2525 rq->ref_count++; 2528 rq->ref_count++;
2526 2529
2527 if (!rq->sense) { 2530 if (!rq->sense) {
2528 memset(sense, 0, sizeof(sense)); 2531 memset(sense, 0, sizeof(sense));
2529 rq->sense = sense; 2532 rq->sense = sense;
2530 rq->sense_len = 0; 2533 rq->sense_len = 0;
2531 } 2534 }
2532 2535
2533 rq->waiting = &wait; 2536 rq->waiting = &wait;
2534 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq); 2537 blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);
2535 wait_for_completion(&wait); 2538 wait_for_completion(&wait);
2536 rq->waiting = NULL; 2539 rq->waiting = NULL;
2537 2540
2538 if (rq->errors) 2541 if (rq->errors)
2539 err = -EIO; 2542 err = -EIO;
2540 2543
2541 return err; 2544 return err;
2542 } 2545 }
2543 2546
2544 EXPORT_SYMBOL(blk_execute_rq); 2547 EXPORT_SYMBOL(blk_execute_rq);
2545 2548
2546 /** 2549 /**
2547 * blkdev_issue_flush - queue a flush 2550 * blkdev_issue_flush - queue a flush
2548 * @bdev: blockdev to issue flush for 2551 * @bdev: blockdev to issue flush for
2549 * @error_sector: error sector 2552 * @error_sector: error sector
2550 * 2553 *
2551 * Description: 2554 * Description:
2552 * Issue a flush for the block device in question. Caller can supply 2555 * Issue a flush for the block device in question. Caller can supply
2553 * room for storing the error offset in case of a flush error, if they 2556 * room for storing the error offset in case of a flush error, if they
2554 * wish to. Caller must run wait_for_completion() on its own. 2557 * wish to. Caller must run wait_for_completion() on its own.
2555 */ 2558 */
2556 int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector) 2559 int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
2557 { 2560 {
2558 request_queue_t *q; 2561 request_queue_t *q;
2559 2562
2560 if (bdev->bd_disk == NULL) 2563 if (bdev->bd_disk == NULL)
2561 return -ENXIO; 2564 return -ENXIO;
2562 2565
2563 q = bdev_get_queue(bdev); 2566 q = bdev_get_queue(bdev);
2564 if (!q) 2567 if (!q)
2565 return -ENXIO; 2568 return -ENXIO;
2566 if (!q->issue_flush_fn) 2569 if (!q->issue_flush_fn)
2567 return -EOPNOTSUPP; 2570 return -EOPNOTSUPP;
2568 2571
2569 return q->issue_flush_fn(q, bdev->bd_disk, error_sector); 2572 return q->issue_flush_fn(q, bdev->bd_disk, error_sector);
2570 } 2573 }
2571 2574
2572 EXPORT_SYMBOL(blkdev_issue_flush); 2575 EXPORT_SYMBOL(blkdev_issue_flush);
2573 2576
2574 static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io) 2577 static void drive_stat_acct(struct request *rq, int nr_sectors, int new_io)
2575 { 2578 {
2576 int rw = rq_data_dir(rq); 2579 int rw = rq_data_dir(rq);
2577 2580
2578 if (!blk_fs_request(rq) || !rq->rq_disk) 2581 if (!blk_fs_request(rq) || !rq->rq_disk)
2579 return; 2582 return;
2580 2583
2581 if (!new_io) { 2584 if (!new_io) {
2582 __disk_stat_inc(rq->rq_disk, merges[rw]); 2585 __disk_stat_inc(rq->rq_disk, merges[rw]);
2583 } else { 2586 } else {
2584 disk_round_stats(rq->rq_disk); 2587 disk_round_stats(rq->rq_disk);
2585 rq->rq_disk->in_flight++; 2588 rq->rq_disk->in_flight++;
2586 } 2589 }
2587 } 2590 }
2588 2591
2589 /* 2592 /*
2590 * add-request adds a request to the linked list. 2593 * add-request adds a request to the linked list.
2591 * queue lock is held and interrupts disabled, as we muck with the 2594 * queue lock is held and interrupts disabled, as we muck with the
2592 * request queue list. 2595 * request queue list.
2593 */ 2596 */
2594 static inline void add_request(request_queue_t * q, struct request * req) 2597 static inline void add_request(request_queue_t * q, struct request * req)
2595 { 2598 {
2596 drive_stat_acct(req, req->nr_sectors, 1); 2599 drive_stat_acct(req, req->nr_sectors, 1);
2597 2600
2598 if (q->activity_fn) 2601 if (q->activity_fn)
2599 q->activity_fn(q->activity_data, rq_data_dir(req)); 2602 q->activity_fn(q->activity_data, rq_data_dir(req));
2600 2603
2601 /* 2604 /*
2602 * elevator indicated where it wants this request to be 2605 * elevator indicated where it wants this request to be
2603 * inserted at elevator_merge time 2606 * inserted at elevator_merge time
2604 */ 2607 */
2605 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0); 2608 __elv_add_request(q, req, ELEVATOR_INSERT_SORT, 0);
2606 } 2609 }
2607 2610
2608 /* 2611 /*
2609 * disk_round_stats() - Round off the performance stats on a struct 2612 * disk_round_stats() - Round off the performance stats on a struct
2610 * disk_stats. 2613 * disk_stats.
2611 * 2614 *
2612 * The average IO queue length and utilisation statistics are maintained 2615 * The average IO queue length and utilisation statistics are maintained
2613 * by observing the current state of the queue length and the amount of 2616 * by observing the current state of the queue length and the amount of
2614 * time it has been in this state for. 2617 * time it has been in this state for.
2615 * 2618 *
2616 * Normally, that accounting is done on IO completion, but that can result 2619 * Normally, that accounting is done on IO completion, but that can result
2617 * in more than a second's worth of IO being accounted for within any one 2620 * in more than a second's worth of IO being accounted for within any one
2618 * second, leading to >100% utilisation. To deal with that, we call this 2621 * second, leading to >100% utilisation. To deal with that, we call this
2619 * function to do a round-off before returning the results when reading 2622 * function to do a round-off before returning the results when reading
2620 * /proc/diskstats. This accounts immediately for all queue usage up to 2623 * /proc/diskstats. This accounts immediately for all queue usage up to
2621 * the current jiffies and restarts the counters again. 2624 * the current jiffies and restarts the counters again.
2622 */ 2625 */
2623 void disk_round_stats(struct gendisk *disk) 2626 void disk_round_stats(struct gendisk *disk)
2624 { 2627 {
2625 unsigned long now = jiffies; 2628 unsigned long now = jiffies;
2626 2629
2627 if (now == disk->stamp) 2630 if (now == disk->stamp)
2628 return; 2631 return;
2629 2632
2630 if (disk->in_flight) { 2633 if (disk->in_flight) {
2631 __disk_stat_add(disk, time_in_queue, 2634 __disk_stat_add(disk, time_in_queue,
2632 disk->in_flight * (now - disk->stamp)); 2635 disk->in_flight * (now - disk->stamp));
2633 __disk_stat_add(disk, io_ticks, (now - disk->stamp)); 2636 __disk_stat_add(disk, io_ticks, (now - disk->stamp));
2634 } 2637 }
2635 disk->stamp = now; 2638 disk->stamp = now;
2636 } 2639 }
2637 2640
2638 EXPORT_SYMBOL_GPL(disk_round_stats); 2641 EXPORT_SYMBOL_GPL(disk_round_stats);
2639 2642
2640 /* 2643 /*
2641 * queue lock must be held 2644 * queue lock must be held
2642 */ 2645 */
2643 void __blk_put_request(request_queue_t *q, struct request *req) 2646 void __blk_put_request(request_queue_t *q, struct request *req)
2644 { 2647 {
2645 struct request_list *rl = req->rl; 2648 struct request_list *rl = req->rl;
2646 2649
2647 if (unlikely(!q)) 2650 if (unlikely(!q))
2648 return; 2651 return;
2649 if (unlikely(--req->ref_count)) 2652 if (unlikely(--req->ref_count))
2650 return; 2653 return;
2651 2654
2652 elv_completed_request(q, req); 2655 elv_completed_request(q, req);
2653 2656
2654 req->rq_status = RQ_INACTIVE; 2657 req->rq_status = RQ_INACTIVE;
2655 req->rl = NULL; 2658 req->rl = NULL;
2656 2659
2657 /* 2660 /*
2658 * Request may not have originated from ll_rw_blk. if not, 2661 * Request may not have originated from ll_rw_blk. if not,
2659 * it didn't come out of our reserved rq pools 2662 * it didn't come out of our reserved rq pools
2660 */ 2663 */
2661 if (rl) { 2664 if (rl) {
2662 int rw = rq_data_dir(req); 2665 int rw = rq_data_dir(req);
2663 int priv = req->flags & REQ_ELVPRIV; 2666 int priv = req->flags & REQ_ELVPRIV;
2664 2667
2665 BUG_ON(!list_empty(&req->queuelist)); 2668 BUG_ON(!list_empty(&req->queuelist));
2666 2669
2667 blk_free_request(q, req); 2670 blk_free_request(q, req);
2668 freed_request(q, rw, priv); 2671 freed_request(q, rw, priv);
2669 } 2672 }
2670 } 2673 }
2671 2674
2672 EXPORT_SYMBOL_GPL(__blk_put_request); 2675 EXPORT_SYMBOL_GPL(__blk_put_request);
2673 2676
2674 void blk_put_request(struct request *req) 2677 void blk_put_request(struct request *req)
2675 { 2678 {
2676 unsigned long flags; 2679 unsigned long flags;
2677 request_queue_t *q = req->q; 2680 request_queue_t *q = req->q;
2678 2681
2679 /* 2682 /*
2680 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the 2683 * Gee, IDE calls in w/ NULL q. Fix IDE and remove the
2681 * following if (q) test. 2684 * following if (q) test.
2682 */ 2685 */
2683 if (q) { 2686 if (q) {
2684 spin_lock_irqsave(q->queue_lock, flags); 2687 spin_lock_irqsave(q->queue_lock, flags);
2685 __blk_put_request(q, req); 2688 __blk_put_request(q, req);
2686 spin_unlock_irqrestore(q->queue_lock, flags); 2689 spin_unlock_irqrestore(q->queue_lock, flags);
2687 } 2690 }
2688 } 2691 }
2689 2692
2690 EXPORT_SYMBOL(blk_put_request); 2693 EXPORT_SYMBOL(blk_put_request);
2691 2694
2692 /** 2695 /**
2693 * blk_end_sync_rq - executes a completion event on a request 2696 * blk_end_sync_rq - executes a completion event on a request
2694 * @rq: request to complete 2697 * @rq: request to complete
2695 * @error: end io status of the request 2698 * @error: end io status of the request
2696 */ 2699 */
2697 void blk_end_sync_rq(struct request *rq, int error) 2700 void blk_end_sync_rq(struct request *rq, int error)
2698 { 2701 {
2699 struct completion *waiting = rq->waiting; 2702 struct completion *waiting = rq->waiting;
2700 2703
2701 rq->waiting = NULL; 2704 rq->waiting = NULL;
2702 __blk_put_request(rq->q, rq); 2705 __blk_put_request(rq->q, rq);
2703 2706
2704 /* 2707 /*
2705 * complete last, if this is a stack request the process (and thus 2708 * complete last, if this is a stack request the process (and thus
2706 * the rq pointer) could be invalid right after this complete() 2709 * the rq pointer) could be invalid right after this complete()
2707 */ 2710 */
2708 complete(waiting); 2711 complete(waiting);
2709 } 2712 }
2710 EXPORT_SYMBOL(blk_end_sync_rq); 2713 EXPORT_SYMBOL(blk_end_sync_rq);
2711 2714
2712 /** 2715 /**
2713 * blk_congestion_wait - wait for a queue to become uncongested 2716 * blk_congestion_wait - wait for a queue to become uncongested
2714 * @rw: READ or WRITE 2717 * @rw: READ or WRITE
2715 * @timeout: timeout in jiffies 2718 * @timeout: timeout in jiffies
2716 * 2719 *
2717 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion. 2720 * Waits for up to @timeout jiffies for a queue (any queue) to exit congestion.
2718 * If no queues are congested then just wait for the next request to be 2721 * If no queues are congested then just wait for the next request to be
2719 * returned. 2722 * returned.
2720 */ 2723 */
2721 long blk_congestion_wait(int rw, long timeout) 2724 long blk_congestion_wait(int rw, long timeout)
2722 { 2725 {
2723 long ret; 2726 long ret;
2724 DEFINE_WAIT(wait); 2727 DEFINE_WAIT(wait);
2725 wait_queue_head_t *wqh = &congestion_wqh[rw]; 2728 wait_queue_head_t *wqh = &congestion_wqh[rw];
2726 2729
2727 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); 2730 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
2728 ret = io_schedule_timeout(timeout); 2731 ret = io_schedule_timeout(timeout);
2729 finish_wait(wqh, &wait); 2732 finish_wait(wqh, &wait);
2730 return ret; 2733 return ret;
2731 } 2734 }
2732 2735
2733 EXPORT_SYMBOL(blk_congestion_wait); 2736 EXPORT_SYMBOL(blk_congestion_wait);
2734 2737
2735 /* 2738 /*
2736 * Has to be called with the request spinlock acquired 2739 * Has to be called with the request spinlock acquired
2737 */ 2740 */
2738 static int attempt_merge(request_queue_t *q, struct request *req, 2741 static int attempt_merge(request_queue_t *q, struct request *req,
2739 struct request *next) 2742 struct request *next)
2740 { 2743 {
2741 if (!rq_mergeable(req) || !rq_mergeable(next)) 2744 if (!rq_mergeable(req) || !rq_mergeable(next))
2742 return 0; 2745 return 0;
2743 2746
2744 /* 2747 /*
2745 * not contigious 2748 * not contigious
2746 */ 2749 */
2747 if (req->sector + req->nr_sectors != next->sector) 2750 if (req->sector + req->nr_sectors != next->sector)
2748 return 0; 2751 return 0;
2749 2752
2750 if (rq_data_dir(req) != rq_data_dir(next) 2753 if (rq_data_dir(req) != rq_data_dir(next)
2751 || req->rq_disk != next->rq_disk 2754 || req->rq_disk != next->rq_disk
2752 || next->waiting || next->special) 2755 || next->waiting || next->special)
2753 return 0; 2756 return 0;
2754 2757
2755 /* 2758 /*
2756 * If we are allowed to merge, then append bio list 2759 * If we are allowed to merge, then append bio list
2757 * from next to rq and release next. merge_requests_fn 2760 * from next to rq and release next. merge_requests_fn
2758 * will have updated segment counts, update sector 2761 * will have updated segment counts, update sector
2759 * counts here. 2762 * counts here.
2760 */ 2763 */
2761 if (!q->merge_requests_fn(q, req, next)) 2764 if (!q->merge_requests_fn(q, req, next))
2762 return 0; 2765 return 0;
2763 2766
2764 /* 2767 /*
2765 * At this point we have either done a back merge 2768 * At this point we have either done a back merge
2766 * or front merge. We need the smaller start_time of 2769 * or front merge. We need the smaller start_time of
2767 * the merged requests to be the current request 2770 * the merged requests to be the current request
2768 * for accounting purposes. 2771 * for accounting purposes.
2769 */ 2772 */
2770 if (time_after(req->start_time, next->start_time)) 2773 if (time_after(req->start_time, next->start_time))
2771 req->start_time = next->start_time; 2774 req->start_time = next->start_time;
2772 2775
2773 req->biotail->bi_next = next->bio; 2776 req->biotail->bi_next = next->bio;
2774 req->biotail = next->biotail; 2777 req->biotail = next->biotail;
2775 2778
2776 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors; 2779 req->nr_sectors = req->hard_nr_sectors += next->hard_nr_sectors;
2777 2780
2778 elv_merge_requests(q, req, next); 2781 elv_merge_requests(q, req, next);
2779 2782
2780 if (req->rq_disk) { 2783 if (req->rq_disk) {
2781 disk_round_stats(req->rq_disk); 2784 disk_round_stats(req->rq_disk);
2782 req->rq_disk->in_flight--; 2785 req->rq_disk->in_flight--;
2783 } 2786 }
2784 2787
2785 req->ioprio = ioprio_best(req->ioprio, next->ioprio); 2788 req->ioprio = ioprio_best(req->ioprio, next->ioprio);
2786 2789
2787 __blk_put_request(q, next); 2790 __blk_put_request(q, next);
2788 return 1; 2791 return 1;
2789 } 2792 }
2790 2793
2791 static inline int attempt_back_merge(request_queue_t *q, struct request *rq) 2794 static inline int attempt_back_merge(request_queue_t *q, struct request *rq)
2792 { 2795 {
2793 struct request *next = elv_latter_request(q, rq); 2796 struct request *next = elv_latter_request(q, rq);
2794 2797
2795 if (next) 2798 if (next)
2796 return attempt_merge(q, rq, next); 2799 return attempt_merge(q, rq, next);
2797 2800
2798 return 0; 2801 return 0;
2799 } 2802 }
2800 2803
2801 static inline int attempt_front_merge(request_queue_t *q, struct request *rq) 2804 static inline int attempt_front_merge(request_queue_t *q, struct request *rq)
2802 { 2805 {
2803 struct request *prev = elv_former_request(q, rq); 2806 struct request *prev = elv_former_request(q, rq);
2804 2807
2805 if (prev) 2808 if (prev)
2806 return attempt_merge(q, prev, rq); 2809 return attempt_merge(q, prev, rq);
2807 2810
2808 return 0; 2811 return 0;
2809 } 2812 }
2810 2813
2811 static void init_request_from_bio(struct request *req, struct bio *bio) 2814 static void init_request_from_bio(struct request *req, struct bio *bio)
2812 { 2815 {
2813 req->flags |= REQ_CMD; 2816 req->flags |= REQ_CMD;
2814 2817
2815 /* 2818 /*
2816 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST) 2819 * inherit FAILFAST from bio (for read-ahead, and explicit FAILFAST)
2817 */ 2820 */
2818 if (bio_rw_ahead(bio) || bio_failfast(bio)) 2821 if (bio_rw_ahead(bio) || bio_failfast(bio))
2819 req->flags |= REQ_FAILFAST; 2822 req->flags |= REQ_FAILFAST;
2820 2823
2821 /* 2824 /*
2822 * REQ_BARRIER implies no merging, but lets make it explicit 2825 * REQ_BARRIER implies no merging, but lets make it explicit
2823 */ 2826 */
2824 if (unlikely(bio_barrier(bio))) 2827 if (unlikely(bio_barrier(bio)))
2825 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE); 2828 req->flags |= (REQ_HARDBARRIER | REQ_NOMERGE);
2826 2829
2827 req->errors = 0; 2830 req->errors = 0;
2828 req->hard_sector = req->sector = bio->bi_sector; 2831 req->hard_sector = req->sector = bio->bi_sector;
2829 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio); 2832 req->hard_nr_sectors = req->nr_sectors = bio_sectors(bio);
2830 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio); 2833 req->current_nr_sectors = req->hard_cur_sectors = bio_cur_sectors(bio);
2831 req->nr_phys_segments = bio_phys_segments(req->q, bio); 2834 req->nr_phys_segments = bio_phys_segments(req->q, bio);
2832 req->nr_hw_segments = bio_hw_segments(req->q, bio); 2835 req->nr_hw_segments = bio_hw_segments(req->q, bio);
2833 req->buffer = bio_data(bio); /* see ->buffer comment above */ 2836 req->buffer = bio_data(bio); /* see ->buffer comment above */
2834 req->waiting = NULL; 2837 req->waiting = NULL;
2835 req->bio = req->biotail = bio; 2838 req->bio = req->biotail = bio;
2836 req->ioprio = bio_prio(bio); 2839 req->ioprio = bio_prio(bio);
2837 req->rq_disk = bio->bi_bdev->bd_disk; 2840 req->rq_disk = bio->bi_bdev->bd_disk;
2838 req->start_time = jiffies; 2841 req->start_time = jiffies;
2839 } 2842 }
2840 2843
2841 static int __make_request(request_queue_t *q, struct bio *bio) 2844 static int __make_request(request_queue_t *q, struct bio *bio)
2842 { 2845 {
2843 struct request *req; 2846 struct request *req;
2844 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync; 2847 int el_ret, rw, nr_sectors, cur_nr_sectors, barrier, err, sync;
2845 unsigned short prio; 2848 unsigned short prio;
2846 sector_t sector; 2849 sector_t sector;
2847 2850
2848 sector = bio->bi_sector; 2851 sector = bio->bi_sector;
2849 nr_sectors = bio_sectors(bio); 2852 nr_sectors = bio_sectors(bio);
2850 cur_nr_sectors = bio_cur_sectors(bio); 2853 cur_nr_sectors = bio_cur_sectors(bio);
2851 prio = bio_prio(bio); 2854 prio = bio_prio(bio);
2852 2855
2853 rw = bio_data_dir(bio); 2856 rw = bio_data_dir(bio);
2854 sync = bio_sync(bio); 2857 sync = bio_sync(bio);
2855 2858
2856 /* 2859 /*
2857 * low level driver can indicate that it wants pages above a 2860 * low level driver can indicate that it wants pages above a
2858 * certain limit bounced to low memory (ie for highmem, or even 2861 * certain limit bounced to low memory (ie for highmem, or even
2859 * ISA dma in theory) 2862 * ISA dma in theory)
2860 */ 2863 */
2861 blk_queue_bounce(q, &bio); 2864 blk_queue_bounce(q, &bio);
2862 2865
2863 spin_lock_prefetch(q->queue_lock); 2866 spin_lock_prefetch(q->queue_lock);
2864 2867
2865 barrier = bio_barrier(bio); 2868 barrier = bio_barrier(bio);
2866 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) { 2869 if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
2867 err = -EOPNOTSUPP; 2870 err = -EOPNOTSUPP;
2868 goto end_io; 2871 goto end_io;
2869 } 2872 }
2870 2873
2871 spin_lock_irq(q->queue_lock); 2874 spin_lock_irq(q->queue_lock);
2872 2875
2873 if (unlikely(barrier) || elv_queue_empty(q)) 2876 if (unlikely(barrier) || elv_queue_empty(q))
2874 goto get_rq; 2877 goto get_rq;
2875 2878
2876 el_ret = elv_merge(q, &req, bio); 2879 el_ret = elv_merge(q, &req, bio);
2877 switch (el_ret) { 2880 switch (el_ret) {
2878 case ELEVATOR_BACK_MERGE: 2881 case ELEVATOR_BACK_MERGE:
2879 BUG_ON(!rq_mergeable(req)); 2882 BUG_ON(!rq_mergeable(req));
2880 2883
2881 if (!q->back_merge_fn(q, req, bio)) 2884 if (!q->back_merge_fn(q, req, bio))
2882 break; 2885 break;
2883 2886
2884 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); 2887 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
2885 2888
2886 req->biotail->bi_next = bio; 2889 req->biotail->bi_next = bio;
2887 req->biotail = bio; 2890 req->biotail = bio;
2888 req->nr_sectors = req->hard_nr_sectors += nr_sectors; 2891 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2889 req->ioprio = ioprio_best(req->ioprio, prio); 2892 req->ioprio = ioprio_best(req->ioprio, prio);
2890 drive_stat_acct(req, nr_sectors, 0); 2893 drive_stat_acct(req, nr_sectors, 0);
2891 if (!attempt_back_merge(q, req)) 2894 if (!attempt_back_merge(q, req))
2892 elv_merged_request(q, req); 2895 elv_merged_request(q, req);
2893 goto out; 2896 goto out;
2894 2897
2895 case ELEVATOR_FRONT_MERGE: 2898 case ELEVATOR_FRONT_MERGE:
2896 BUG_ON(!rq_mergeable(req)); 2899 BUG_ON(!rq_mergeable(req));
2897 2900
2898 if (!q->front_merge_fn(q, req, bio)) 2901 if (!q->front_merge_fn(q, req, bio))
2899 break; 2902 break;
2900 2903
2901 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); 2904 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
2902 2905
2903 bio->bi_next = req->bio; 2906 bio->bi_next = req->bio;
2904 req->bio = bio; 2907 req->bio = bio;
2905 2908
2906 /* 2909 /*
2907 * may not be valid. if the low level driver said 2910 * may not be valid. if the low level driver said
2908 * it didn't need a bounce buffer then it better 2911 * it didn't need a bounce buffer then it better
2909 * not touch req->buffer either... 2912 * not touch req->buffer either...
2910 */ 2913 */
2911 req->buffer = bio_data(bio); 2914 req->buffer = bio_data(bio);
2912 req->current_nr_sectors = cur_nr_sectors; 2915 req->current_nr_sectors = cur_nr_sectors;
2913 req->hard_cur_sectors = cur_nr_sectors; 2916 req->hard_cur_sectors = cur_nr_sectors;
2914 req->sector = req->hard_sector = sector; 2917 req->sector = req->hard_sector = sector;
2915 req->nr_sectors = req->hard_nr_sectors += nr_sectors; 2918 req->nr_sectors = req->hard_nr_sectors += nr_sectors;
2916 req->ioprio = ioprio_best(req->ioprio, prio); 2919 req->ioprio = ioprio_best(req->ioprio, prio);
2917 drive_stat_acct(req, nr_sectors, 0); 2920 drive_stat_acct(req, nr_sectors, 0);
2918 if (!attempt_front_merge(q, req)) 2921 if (!attempt_front_merge(q, req))
2919 elv_merged_request(q, req); 2922 elv_merged_request(q, req);
2920 goto out; 2923 goto out;
2921 2924
2922 /* ELV_NO_MERGE: elevator says don't/can't merge. */ 2925 /* ELV_NO_MERGE: elevator says don't/can't merge. */
2923 default: 2926 default:
2924 ; 2927 ;
2925 } 2928 }
2926 2929
2927 get_rq: 2930 get_rq:
2928 /* 2931 /*
2929 * Grab a free request. This is might sleep but can not fail. 2932 * Grab a free request. This is might sleep but can not fail.
2930 * Returns with the queue unlocked. 2933 * Returns with the queue unlocked.
2931 */ 2934 */
2932 req = get_request_wait(q, rw, bio); 2935 req = get_request_wait(q, rw, bio);
2933 2936
2934 /* 2937 /*
2935 * After dropping the lock and possibly sleeping here, our request 2938 * After dropping the lock and possibly sleeping here, our request
2936 * may now be mergeable after it had proven unmergeable (above). 2939 * may now be mergeable after it had proven unmergeable (above).
2937 * We don't worry about that case for efficiency. It won't happen 2940 * We don't worry about that case for efficiency. It won't happen
2938 * often, and the elevators are able to handle it. 2941 * often, and the elevators are able to handle it.
2939 */ 2942 */
2940 init_request_from_bio(req, bio); 2943 init_request_from_bio(req, bio);
2941 2944
2942 spin_lock_irq(q->queue_lock); 2945 spin_lock_irq(q->queue_lock);
2943 if (elv_queue_empty(q)) 2946 if (elv_queue_empty(q))
2944 blk_plug_device(q); 2947 blk_plug_device(q);
2945 add_request(q, req); 2948 add_request(q, req);
2946 out: 2949 out:
2947 if (sync) 2950 if (sync)
2948 __generic_unplug_device(q); 2951 __generic_unplug_device(q);
2949 2952
2950 spin_unlock_irq(q->queue_lock); 2953 spin_unlock_irq(q->queue_lock);
2951 return 0; 2954 return 0;
2952 2955
2953 end_io: 2956 end_io:
2954 bio_endio(bio, nr_sectors << 9, err); 2957 bio_endio(bio, nr_sectors << 9, err);
2955 return 0; 2958 return 0;
2956 } 2959 }
2957 2960
2958 /* 2961 /*
2959 * If bio->bi_dev is a partition, remap the location 2962 * If bio->bi_dev is a partition, remap the location
2960 */ 2963 */
2961 static inline void blk_partition_remap(struct bio *bio) 2964 static inline void blk_partition_remap(struct bio *bio)
2962 { 2965 {
2963 struct block_device *bdev = bio->bi_bdev; 2966 struct block_device *bdev = bio->bi_bdev;
2964 2967
2965 if (bdev != bdev->bd_contains) { 2968 if (bdev != bdev->bd_contains) {
2966 struct hd_struct *p = bdev->bd_part; 2969 struct hd_struct *p = bdev->bd_part;
2967 const int rw = bio_data_dir(bio); 2970 const int rw = bio_data_dir(bio);
2968 2971
2969 p->sectors[rw] += bio_sectors(bio); 2972 p->sectors[rw] += bio_sectors(bio);
2970 p->ios[rw]++; 2973 p->ios[rw]++;
2971 2974
2972 bio->bi_sector += p->start_sect; 2975 bio->bi_sector += p->start_sect;
2973 bio->bi_bdev = bdev->bd_contains; 2976 bio->bi_bdev = bdev->bd_contains;
2974 } 2977 }
2975 } 2978 }
2976 2979
2977 static void handle_bad_sector(struct bio *bio) 2980 static void handle_bad_sector(struct bio *bio)
2978 { 2981 {
2979 char b[BDEVNAME_SIZE]; 2982 char b[BDEVNAME_SIZE];
2980 2983
2981 printk(KERN_INFO "attempt to access beyond end of device\n"); 2984 printk(KERN_INFO "attempt to access beyond end of device\n");
2982 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", 2985 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
2983 bdevname(bio->bi_bdev, b), 2986 bdevname(bio->bi_bdev, b),
2984 bio->bi_rw, 2987 bio->bi_rw,
2985 (unsigned long long)bio->bi_sector + bio_sectors(bio), 2988 (unsigned long long)bio->bi_sector + bio_sectors(bio),
2986 (long long)(bio->bi_bdev->bd_inode->i_size >> 9)); 2989 (long long)(bio->bi_bdev->bd_inode->i_size >> 9));
2987 2990
2988 set_bit(BIO_EOF, &bio->bi_flags); 2991 set_bit(BIO_EOF, &bio->bi_flags);
2989 } 2992 }
2990 2993
2991 /** 2994 /**
2992 * generic_make_request: hand a buffer to its device driver for I/O 2995 * generic_make_request: hand a buffer to its device driver for I/O
2993 * @bio: The bio describing the location in memory and on the device. 2996 * @bio: The bio describing the location in memory and on the device.
2994 * 2997 *
2995 * generic_make_request() is used to make I/O requests of block 2998 * generic_make_request() is used to make I/O requests of block
2996 * devices. It is passed a &struct bio, which describes the I/O that needs 2999 * devices. It is passed a &struct bio, which describes the I/O that needs
2997 * to be done. 3000 * to be done.
2998 * 3001 *
2999 * generic_make_request() does not return any status. The 3002 * generic_make_request() does not return any status. The
3000 * success/failure status of the request, along with notification of 3003 * success/failure status of the request, along with notification of
3001 * completion, is delivered asynchronously through the bio->bi_end_io 3004 * completion, is delivered asynchronously through the bio->bi_end_io
3002 * function described (one day) else where. 3005 * function described (one day) else where.
3003 * 3006 *
3004 * The caller of generic_make_request must make sure that bi_io_vec 3007 * The caller of generic_make_request must make sure that bi_io_vec
3005 * are set to describe the memory buffer, and that bi_dev and bi_sector are 3008 * are set to describe the memory buffer, and that bi_dev and bi_sector are
3006 * set to describe the device address, and the 3009 * set to describe the device address, and the
3007 * bi_end_io and optionally bi_private are set to describe how 3010 * bi_end_io and optionally bi_private are set to describe how
3008 * completion notification should be signaled. 3011 * completion notification should be signaled.
3009 * 3012 *
3010 * generic_make_request and the drivers it calls may use bi_next if this 3013 * generic_make_request and the drivers it calls may use bi_next if this
3011 * bio happens to be merged with someone else, and may change bi_dev and 3014 * bio happens to be merged with someone else, and may change bi_dev and
3012 * bi_sector for remaps as it sees fit. So the values of these fields 3015 * bi_sector for remaps as it sees fit. So the values of these fields
3013 * should NOT be depended on after the call to generic_make_request. 3016 * should NOT be depended on after the call to generic_make_request.
3014 */ 3017 */
3015 void generic_make_request(struct bio *bio) 3018 void generic_make_request(struct bio *bio)
3016 { 3019 {
3017 request_queue_t *q; 3020 request_queue_t *q;
3018 sector_t maxsector; 3021 sector_t maxsector;
3019 int ret, nr_sectors = bio_sectors(bio); 3022 int ret, nr_sectors = bio_sectors(bio);
3020 dev_t old_dev; 3023 dev_t old_dev;
3021 3024
3022 might_sleep(); 3025 might_sleep();
3023 /* Test device or partition size, when known. */ 3026 /* Test device or partition size, when known. */
3024 maxsector = bio->bi_bdev->bd_inode->i_size >> 9; 3027 maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
3025 if (maxsector) { 3028 if (maxsector) {
3026 sector_t sector = bio->bi_sector; 3029 sector_t sector = bio->bi_sector;
3027 3030
3028 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { 3031 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
3029 /* 3032 /*
3030 * This may well happen - the kernel calls bread() 3033 * This may well happen - the kernel calls bread()
3031 * without checking the size of the device, e.g., when 3034 * without checking the size of the device, e.g., when
3032 * mounting a device. 3035 * mounting a device.
3033 */ 3036 */
3034 handle_bad_sector(bio); 3037 handle_bad_sector(bio);
3035 goto end_io; 3038 goto end_io;
3036 } 3039 }
3037 } 3040 }
3038 3041
3039 /* 3042 /*
3040 * Resolve the mapping until finished. (drivers are 3043 * Resolve the mapping until finished. (drivers are
3041 * still free to implement/resolve their own stacking 3044 * still free to implement/resolve their own stacking
3042 * by explicitly returning 0) 3045 * by explicitly returning 0)
3043 * 3046 *
3044 * NOTE: we don't repeat the blk_size check for each new device. 3047 * NOTE: we don't repeat the blk_size check for each new device.
3045 * Stacking drivers are expected to know what they are doing. 3048 * Stacking drivers are expected to know what they are doing.
3046 */ 3049 */
3047 maxsector = -1; 3050 maxsector = -1;
3048 old_dev = 0; 3051 old_dev = 0;
3049 do { 3052 do {
3050 char b[BDEVNAME_SIZE]; 3053 char b[BDEVNAME_SIZE];
3051 3054
3052 q = bdev_get_queue(bio->bi_bdev); 3055 q = bdev_get_queue(bio->bi_bdev);
3053 if (!q) { 3056 if (!q) {
3054 printk(KERN_ERR 3057 printk(KERN_ERR
3055 "generic_make_request: Trying to access " 3058 "generic_make_request: Trying to access "
3056 "nonexistent block-device %s (%Lu)\n", 3059 "nonexistent block-device %s (%Lu)\n",
3057 bdevname(bio->bi_bdev, b), 3060 bdevname(bio->bi_bdev, b),
3058 (long long) bio->bi_sector); 3061 (long long) bio->bi_sector);
3059 end_io: 3062 end_io:
3060 bio_endio(bio, bio->bi_size, -EIO); 3063 bio_endio(bio, bio->bi_size, -EIO);
3061 break; 3064 break;
3062 } 3065 }
3063 3066
3064 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) { 3067 if (unlikely(bio_sectors(bio) > q->max_hw_sectors)) {
3065 printk("bio too big device %s (%u > %u)\n", 3068 printk("bio too big device %s (%u > %u)\n",
3066 bdevname(bio->bi_bdev, b), 3069 bdevname(bio->bi_bdev, b),
3067 bio_sectors(bio), 3070 bio_sectors(bio),
3068 q->max_hw_sectors); 3071 q->max_hw_sectors);
3069 goto end_io; 3072 goto end_io;
3070 } 3073 }
3071 3074
3072 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags))) 3075 if (unlikely(test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)))
3073 goto end_io; 3076 goto end_io;
3074 3077
3075 /* 3078 /*
3076 * If this device has partitions, remap block n 3079 * If this device has partitions, remap block n
3077 * of partition p to block n+start(p) of the disk. 3080 * of partition p to block n+start(p) of the disk.
3078 */ 3081 */
3079 blk_partition_remap(bio); 3082 blk_partition_remap(bio);
3080 3083
3081 if (maxsector != -1) 3084 if (maxsector != -1)
3082 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector, 3085 blk_add_trace_remap(q, bio, old_dev, bio->bi_sector,
3083 maxsector); 3086 maxsector);
3084 3087
3085 blk_add_trace_bio(q, bio, BLK_TA_QUEUE); 3088 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
3086 3089
3087 maxsector = bio->bi_sector; 3090 maxsector = bio->bi_sector;
3088 old_dev = bio->bi_bdev->bd_dev; 3091 old_dev = bio->bi_bdev->bd_dev;
3089 3092
3090 ret = q->make_request_fn(q, bio); 3093 ret = q->make_request_fn(q, bio);
3091 } while (ret); 3094 } while (ret);
3092 } 3095 }
3093 3096
3094 EXPORT_SYMBOL(generic_make_request); 3097 EXPORT_SYMBOL(generic_make_request);
3095 3098
3096 /** 3099 /**
3097 * submit_bio: submit a bio to the block device layer for I/O 3100 * submit_bio: submit a bio to the block device layer for I/O
3098 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 3101 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
3099 * @bio: The &struct bio which describes the I/O 3102 * @bio: The &struct bio which describes the I/O
3100 * 3103 *
3101 * submit_bio() is very similar in purpose to generic_make_request(), and 3104 * submit_bio() is very similar in purpose to generic_make_request(), and
3102 * uses that function to do most of the work. Both are fairly rough 3105 * uses that function to do most of the work. Both are fairly rough
3103 * interfaces, @bio must be presetup and ready for I/O. 3106 * interfaces, @bio must be presetup and ready for I/O.
3104 * 3107 *
3105 */ 3108 */
3106 void submit_bio(int rw, struct bio *bio) 3109 void submit_bio(int rw, struct bio *bio)
3107 { 3110 {
3108 int count = bio_sectors(bio); 3111 int count = bio_sectors(bio);
3109 3112
3110 BIO_BUG_ON(!bio->bi_size); 3113 BIO_BUG_ON(!bio->bi_size);
3111 BIO_BUG_ON(!bio->bi_io_vec); 3114 BIO_BUG_ON(!bio->bi_io_vec);
3112 bio->bi_rw |= rw; 3115 bio->bi_rw |= rw;
3113 if (rw & WRITE) 3116 if (rw & WRITE)
3114 mod_page_state(pgpgout, count); 3117 mod_page_state(pgpgout, count);
3115 else 3118 else
3116 mod_page_state(pgpgin, count); 3119 mod_page_state(pgpgin, count);
3117 3120
3118 if (unlikely(block_dump)) { 3121 if (unlikely(block_dump)) {
3119 char b[BDEVNAME_SIZE]; 3122 char b[BDEVNAME_SIZE];
3120 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n", 3123 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s\n",
3121 current->comm, current->pid, 3124 current->comm, current->pid,
3122 (rw & WRITE) ? "WRITE" : "READ", 3125 (rw & WRITE) ? "WRITE" : "READ",
3123 (unsigned long long)bio->bi_sector, 3126 (unsigned long long)bio->bi_sector,
3124 bdevname(bio->bi_bdev,b)); 3127 bdevname(bio->bi_bdev,b));
3125 } 3128 }
3126 3129
3127 generic_make_request(bio); 3130 generic_make_request(bio);
3128 } 3131 }
3129 3132
3130 EXPORT_SYMBOL(submit_bio); 3133 EXPORT_SYMBOL(submit_bio);
3131 3134
3132 static void blk_recalc_rq_segments(struct request *rq) 3135 static void blk_recalc_rq_segments(struct request *rq)
3133 { 3136 {
3134 struct bio *bio, *prevbio = NULL; 3137 struct bio *bio, *prevbio = NULL;
3135 int nr_phys_segs, nr_hw_segs; 3138 int nr_phys_segs, nr_hw_segs;
3136 unsigned int phys_size, hw_size; 3139 unsigned int phys_size, hw_size;
3137 request_queue_t *q = rq->q; 3140 request_queue_t *q = rq->q;
3138 3141
3139 if (!rq->bio) 3142 if (!rq->bio)
3140 return; 3143 return;
3141 3144
3142 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0; 3145 phys_size = hw_size = nr_phys_segs = nr_hw_segs = 0;
3143 rq_for_each_bio(bio, rq) { 3146 rq_for_each_bio(bio, rq) {
3144 /* Force bio hw/phys segs to be recalculated. */ 3147 /* Force bio hw/phys segs to be recalculated. */
3145 bio->bi_flags &= ~(1 << BIO_SEG_VALID); 3148 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
3146 3149
3147 nr_phys_segs += bio_phys_segments(q, bio); 3150 nr_phys_segs += bio_phys_segments(q, bio);
3148 nr_hw_segs += bio_hw_segments(q, bio); 3151 nr_hw_segs += bio_hw_segments(q, bio);
3149 if (prevbio) { 3152 if (prevbio) {
3150 int pseg = phys_size + prevbio->bi_size + bio->bi_size; 3153 int pseg = phys_size + prevbio->bi_size + bio->bi_size;
3151 int hseg = hw_size + prevbio->bi_size + bio->bi_size; 3154 int hseg = hw_size + prevbio->bi_size + bio->bi_size;
3152 3155
3153 if (blk_phys_contig_segment(q, prevbio, bio) && 3156 if (blk_phys_contig_segment(q, prevbio, bio) &&
3154 pseg <= q->max_segment_size) { 3157 pseg <= q->max_segment_size) {
3155 nr_phys_segs--; 3158 nr_phys_segs--;
3156 phys_size += prevbio->bi_size + bio->bi_size; 3159 phys_size += prevbio->bi_size + bio->bi_size;
3157 } else 3160 } else
3158 phys_size = 0; 3161 phys_size = 0;
3159 3162
3160 if (blk_hw_contig_segment(q, prevbio, bio) && 3163 if (blk_hw_contig_segment(q, prevbio, bio) &&
3161 hseg <= q->max_segment_size) { 3164 hseg <= q->max_segment_size) {
3162 nr_hw_segs--; 3165 nr_hw_segs--;
3163 hw_size += prevbio->bi_size + bio->bi_size; 3166 hw_size += prevbio->bi_size + bio->bi_size;
3164 } else 3167 } else
3165 hw_size = 0; 3168 hw_size = 0;
3166 } 3169 }
3167 prevbio = bio; 3170 prevbio = bio;
3168 } 3171 }
3169 3172
3170 rq->nr_phys_segments = nr_phys_segs; 3173 rq->nr_phys_segments = nr_phys_segs;
3171 rq->nr_hw_segments = nr_hw_segs; 3174 rq->nr_hw_segments = nr_hw_segs;
3172 } 3175 }
3173 3176
3174 static void blk_recalc_rq_sectors(struct request *rq, int nsect) 3177 static void blk_recalc_rq_sectors(struct request *rq, int nsect)
3175 { 3178 {
3176 if (blk_fs_request(rq)) { 3179 if (blk_fs_request(rq)) {
3177 rq->hard_sector += nsect; 3180 rq->hard_sector += nsect;
3178 rq->hard_nr_sectors -= nsect; 3181 rq->hard_nr_sectors -= nsect;
3179 3182
3180 /* 3183 /*
3181 * Move the I/O submission pointers ahead if required. 3184 * Move the I/O submission pointers ahead if required.
3182 */ 3185 */
3183 if ((rq->nr_sectors >= rq->hard_nr_sectors) && 3186 if ((rq->nr_sectors >= rq->hard_nr_sectors) &&
3184 (rq->sector <= rq->hard_sector)) { 3187 (rq->sector <= rq->hard_sector)) {
3185 rq->sector = rq->hard_sector; 3188 rq->sector = rq->hard_sector;
3186 rq->nr_sectors = rq->hard_nr_sectors; 3189 rq->nr_sectors = rq->hard_nr_sectors;
3187 rq->hard_cur_sectors = bio_cur_sectors(rq->bio); 3190 rq->hard_cur_sectors = bio_cur_sectors(rq->bio);
3188 rq->current_nr_sectors = rq->hard_cur_sectors; 3191 rq->current_nr_sectors = rq->hard_cur_sectors;
3189 rq->buffer = bio_data(rq->bio); 3192 rq->buffer = bio_data(rq->bio);
3190 } 3193 }
3191 3194
3192 /* 3195 /*
3193 * if total number of sectors is less than the first segment 3196 * if total number of sectors is less than the first segment
3194 * size, something has gone terribly wrong 3197 * size, something has gone terribly wrong
3195 */ 3198 */
3196 if (rq->nr_sectors < rq->current_nr_sectors) { 3199 if (rq->nr_sectors < rq->current_nr_sectors) {
3197 printk("blk: request botched\n"); 3200 printk("blk: request botched\n");
3198 rq->nr_sectors = rq->current_nr_sectors; 3201 rq->nr_sectors = rq->current_nr_sectors;
3199 } 3202 }
3200 } 3203 }
3201 } 3204 }
3202 3205
3203 static int __end_that_request_first(struct request *req, int uptodate, 3206 static int __end_that_request_first(struct request *req, int uptodate,
3204 int nr_bytes) 3207 int nr_bytes)
3205 { 3208 {
3206 int total_bytes, bio_nbytes, error, next_idx = 0; 3209 int total_bytes, bio_nbytes, error, next_idx = 0;
3207 struct bio *bio; 3210 struct bio *bio;
3208 3211
3209 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE); 3212 blk_add_trace_rq(req->q, req, BLK_TA_COMPLETE);
3210 3213
3211 /* 3214 /*
3212 * extend uptodate bool to allow < 0 value to be direct io error 3215 * extend uptodate bool to allow < 0 value to be direct io error
3213 */ 3216 */
3214 error = 0; 3217 error = 0;
3215 if (end_io_error(uptodate)) 3218 if (end_io_error(uptodate))
3216 error = !uptodate ? -EIO : uptodate; 3219 error = !uptodate ? -EIO : uptodate;
3217 3220
3218 /* 3221 /*
3219 * for a REQ_BLOCK_PC request, we want to carry any eventual 3222 * for a REQ_BLOCK_PC request, we want to carry any eventual
3220 * sense key with us all the way through 3223 * sense key with us all the way through
3221 */ 3224 */
3222 if (!blk_pc_request(req)) 3225 if (!blk_pc_request(req))
3223 req->errors = 0; 3226 req->errors = 0;
3224 3227
3225 if (!uptodate) { 3228 if (!uptodate) {
3226 if (blk_fs_request(req) && !(req->flags & REQ_QUIET)) 3229 if (blk_fs_request(req) && !(req->flags & REQ_QUIET))
3227 printk("end_request: I/O error, dev %s, sector %llu\n", 3230 printk("end_request: I/O error, dev %s, sector %llu\n",
3228 req->rq_disk ? req->rq_disk->disk_name : "?", 3231 req->rq_disk ? req->rq_disk->disk_name : "?",
3229 (unsigned long long)req->sector); 3232 (unsigned long long)req->sector);
3230 } 3233 }
3231 3234
3232 if (blk_fs_request(req) && req->rq_disk) { 3235 if (blk_fs_request(req) && req->rq_disk) {
3233 const int rw = rq_data_dir(req); 3236 const int rw = rq_data_dir(req);
3234 3237
3235 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9); 3238 disk_stat_add(req->rq_disk, sectors[rw], nr_bytes >> 9);
3236 } 3239 }
3237 3240
3238 total_bytes = bio_nbytes = 0; 3241 total_bytes = bio_nbytes = 0;
3239 while ((bio = req->bio) != NULL) { 3242 while ((bio = req->bio) != NULL) {
3240 int nbytes; 3243 int nbytes;
3241 3244
3242 if (nr_bytes >= bio->bi_size) { 3245 if (nr_bytes >= bio->bi_size) {
3243 req->bio = bio->bi_next; 3246 req->bio = bio->bi_next;
3244 nbytes = bio->bi_size; 3247 nbytes = bio->bi_size;
3245 if (!ordered_bio_endio(req, bio, nbytes, error)) 3248 if (!ordered_bio_endio(req, bio, nbytes, error))
3246 bio_endio(bio, nbytes, error); 3249 bio_endio(bio, nbytes, error);
3247 next_idx = 0; 3250 next_idx = 0;
3248 bio_nbytes = 0; 3251 bio_nbytes = 0;
3249 } else { 3252 } else {
3250 int idx = bio->bi_idx + next_idx; 3253 int idx = bio->bi_idx + next_idx;
3251 3254
3252 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) { 3255 if (unlikely(bio->bi_idx >= bio->bi_vcnt)) {
3253 blk_dump_rq_flags(req, "__end_that"); 3256 blk_dump_rq_flags(req, "__end_that");
3254 printk("%s: bio idx %d >= vcnt %d\n", 3257 printk("%s: bio idx %d >= vcnt %d\n",
3255 __FUNCTION__, 3258 __FUNCTION__,
3256 bio->bi_idx, bio->bi_vcnt); 3259 bio->bi_idx, bio->bi_vcnt);
3257 break; 3260 break;
3258 } 3261 }
3259 3262
3260 nbytes = bio_iovec_idx(bio, idx)->bv_len; 3263 nbytes = bio_iovec_idx(bio, idx)->bv_len;
3261 BIO_BUG_ON(nbytes > bio->bi_size); 3264 BIO_BUG_ON(nbytes > bio->bi_size);
3262 3265
3263 /* 3266 /*
3264 * not a complete bvec done 3267 * not a complete bvec done
3265 */ 3268 */
3266 if (unlikely(nbytes > nr_bytes)) { 3269 if (unlikely(nbytes > nr_bytes)) {
3267 bio_nbytes += nr_bytes; 3270 bio_nbytes += nr_bytes;
3268 total_bytes += nr_bytes; 3271 total_bytes += nr_bytes;
3269 break; 3272 break;
3270 } 3273 }
3271 3274
3272 /* 3275 /*
3273 * advance to the next vector 3276 * advance to the next vector
3274 */ 3277 */
3275 next_idx++; 3278 next_idx++;
3276 bio_nbytes += nbytes; 3279 bio_nbytes += nbytes;
3277 } 3280 }
3278 3281
3279 total_bytes += nbytes; 3282 total_bytes += nbytes;
3280 nr_bytes -= nbytes; 3283 nr_bytes -= nbytes;
3281 3284
3282 if ((bio = req->bio)) { 3285 if ((bio = req->bio)) {
3283 /* 3286 /*
3284 * end more in this run, or just return 'not-done' 3287 * end more in this run, or just return 'not-done'
3285 */ 3288 */
3286 if (unlikely(nr_bytes <= 0)) 3289 if (unlikely(nr_bytes <= 0))
3287 break; 3290 break;
3288 } 3291 }
3289 } 3292 }
3290 3293
3291 /* 3294 /*
3292 * completely done 3295 * completely done
3293 */ 3296 */
3294 if (!req->bio) 3297 if (!req->bio)
3295 return 0; 3298 return 0;
3296 3299
3297 /* 3300 /*
3298 * if the request wasn't completed, update state 3301 * if the request wasn't completed, update state
3299 */ 3302 */
3300 if (bio_nbytes) { 3303 if (bio_nbytes) {
3301 if (!ordered_bio_endio(req, bio, bio_nbytes, error)) 3304 if (!ordered_bio_endio(req, bio, bio_nbytes, error))
3302 bio_endio(bio, bio_nbytes, error); 3305 bio_endio(bio, bio_nbytes, error);
3303 bio->bi_idx += next_idx; 3306 bio->bi_idx += next_idx;
3304 bio_iovec(bio)->bv_offset += nr_bytes; 3307 bio_iovec(bio)->bv_offset += nr_bytes;
3305 bio_iovec(bio)->bv_len -= nr_bytes; 3308 bio_iovec(bio)->bv_len -= nr_bytes;
3306 } 3309 }
3307 3310
3308 blk_recalc_rq_sectors(req, total_bytes >> 9); 3311 blk_recalc_rq_sectors(req, total_bytes >> 9);
3309 blk_recalc_rq_segments(req); 3312 blk_recalc_rq_segments(req);
3310 return 1; 3313 return 1;
3311 } 3314 }
3312 3315
3313 /** 3316 /**
3314 * end_that_request_first - end I/O on a request 3317 * end_that_request_first - end I/O on a request
3315 * @req: the request being processed 3318 * @req: the request being processed
3316 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error 3319 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3317 * @nr_sectors: number of sectors to end I/O on 3320 * @nr_sectors: number of sectors to end I/O on
3318 * 3321 *
3319 * Description: 3322 * Description:
3320 * Ends I/O on a number of sectors attached to @req, and sets it up 3323 * Ends I/O on a number of sectors attached to @req, and sets it up
3321 * for the next range of segments (if any) in the cluster. 3324 * for the next range of segments (if any) in the cluster.
3322 * 3325 *
3323 * Return: 3326 * Return:
3324 * 0 - we are done with this request, call end_that_request_last() 3327 * 0 - we are done with this request, call end_that_request_last()
3325 * 1 - still buffers pending for this request 3328 * 1 - still buffers pending for this request
3326 **/ 3329 **/
3327 int end_that_request_first(struct request *req, int uptodate, int nr_sectors) 3330 int end_that_request_first(struct request *req, int uptodate, int nr_sectors)
3328 { 3331 {
3329 return __end_that_request_first(req, uptodate, nr_sectors << 9); 3332 return __end_that_request_first(req, uptodate, nr_sectors << 9);
3330 } 3333 }
3331 3334
3332 EXPORT_SYMBOL(end_that_request_first); 3335 EXPORT_SYMBOL(end_that_request_first);
3333 3336
3334 /** 3337 /**
3335 * end_that_request_chunk - end I/O on a request 3338 * end_that_request_chunk - end I/O on a request
3336 * @req: the request being processed 3339 * @req: the request being processed
3337 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error 3340 * @uptodate: 1 for success, 0 for I/O error, < 0 for specific error
3338 * @nr_bytes: number of bytes to complete 3341 * @nr_bytes: number of bytes to complete
3339 * 3342 *
3340 * Description: 3343 * Description:
3341 * Ends I/O on a number of bytes attached to @req, and sets it up 3344 * Ends I/O on a number of bytes attached to @req, and sets it up
3342 * for the next range of segments (if any). Like end_that_request_first(), 3345 * for the next range of segments (if any). Like end_that_request_first(),
3343 * but deals with bytes instead of sectors. 3346 * but deals with bytes instead of sectors.
3344 * 3347 *
3345 * Return: 3348 * Return:
3346 * 0 - we are done with this request, call end_that_request_last() 3349 * 0 - we are done with this request, call end_that_request_last()
3347 * 1 - still buffers pending for this request 3350 * 1 - still buffers pending for this request
3348 **/ 3351 **/
3349 int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes) 3352 int end_that_request_chunk(struct request *req, int uptodate, int nr_bytes)
3350 { 3353 {
3351 return __end_that_request_first(req, uptodate, nr_bytes); 3354 return __end_that_request_first(req, uptodate, nr_bytes);
3352 } 3355 }
3353 3356
3354 EXPORT_SYMBOL(end_that_request_chunk); 3357 EXPORT_SYMBOL(end_that_request_chunk);
3355 3358
3356 /* 3359 /*
3357 * splice the completion data to a local structure and hand off to 3360 * splice the completion data to a local structure and hand off to
3358 * process_completion_queue() to complete the requests 3361 * process_completion_queue() to complete the requests
3359 */ 3362 */
3360 static void blk_done_softirq(struct softirq_action *h) 3363 static void blk_done_softirq(struct softirq_action *h)
3361 { 3364 {
3362 struct list_head *cpu_list, local_list; 3365 struct list_head *cpu_list, local_list;
3363 3366
3364 local_irq_disable(); 3367 local_irq_disable();
3365 cpu_list = &__get_cpu_var(blk_cpu_done); 3368 cpu_list = &__get_cpu_var(blk_cpu_done);
3366 list_replace_init(cpu_list, &local_list); 3369 list_replace_init(cpu_list, &local_list);
3367 local_irq_enable(); 3370 local_irq_enable();
3368 3371
3369 while (!list_empty(&local_list)) { 3372 while (!list_empty(&local_list)) {
3370 struct request *rq = list_entry(local_list.next, struct request, donelist); 3373 struct request *rq = list_entry(local_list.next, struct request, donelist);
3371 3374
3372 list_del_init(&rq->donelist); 3375 list_del_init(&rq->donelist);
3373 rq->q->softirq_done_fn(rq); 3376 rq->q->softirq_done_fn(rq);
3374 } 3377 }
3375 } 3378 }
3376 3379
3377 #ifdef CONFIG_HOTPLUG_CPU 3380 #ifdef CONFIG_HOTPLUG_CPU
3378 3381
3379 static int blk_cpu_notify(struct notifier_block *self, unsigned long action, 3382 static int blk_cpu_notify(struct notifier_block *self, unsigned long action,
3380 void *hcpu) 3383 void *hcpu)
3381 { 3384 {
3382 /* 3385 /*
3383 * If a CPU goes away, splice its entries to the current CPU 3386 * If a CPU goes away, splice its entries to the current CPU
3384 * and trigger a run of the softirq 3387 * and trigger a run of the softirq
3385 */ 3388 */
3386 if (action == CPU_DEAD) { 3389 if (action == CPU_DEAD) {
3387 int cpu = (unsigned long) hcpu; 3390 int cpu = (unsigned long) hcpu;
3388 3391
3389 local_irq_disable(); 3392 local_irq_disable();
3390 list_splice_init(&per_cpu(blk_cpu_done, cpu), 3393 list_splice_init(&per_cpu(blk_cpu_done, cpu),
3391 &__get_cpu_var(blk_cpu_done)); 3394 &__get_cpu_var(blk_cpu_done));
3392 raise_softirq_irqoff(BLOCK_SOFTIRQ); 3395 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3393 local_irq_enable(); 3396 local_irq_enable();
3394 } 3397 }
3395 3398
3396 return NOTIFY_OK; 3399 return NOTIFY_OK;
3397 } 3400 }
3398 3401
3399 3402
3400 static struct notifier_block blk_cpu_notifier = { 3403 static struct notifier_block blk_cpu_notifier = {
3401 .notifier_call = blk_cpu_notify, 3404 .notifier_call = blk_cpu_notify,
3402 }; 3405 };
3403 3406
3404 #endif /* CONFIG_HOTPLUG_CPU */ 3407 #endif /* CONFIG_HOTPLUG_CPU */
3405 3408
3406 /** 3409 /**
3407 * blk_complete_request - end I/O on a request 3410 * blk_complete_request - end I/O on a request
3408 * @req: the request being processed 3411 * @req: the request being processed
3409 * 3412 *
3410 * Description: 3413 * Description:
3411 * Ends all I/O on a request. It does not handle partial completions, 3414 * Ends all I/O on a request. It does not handle partial completions,
3412 * unless the driver actually implements this in its completionc callback 3415 * unless the driver actually implements this in its completionc callback
3413 * through requeueing. Theh actual completion happens out-of-order, 3416 * through requeueing. Theh actual completion happens out-of-order,
3414 * through a softirq handler. The user must have registered a completion 3417 * through a softirq handler. The user must have registered a completion
3415 * callback through blk_queue_softirq_done(). 3418 * callback through blk_queue_softirq_done().
3416 **/ 3419 **/
3417 3420
3418 void blk_complete_request(struct request *req) 3421 void blk_complete_request(struct request *req)
3419 { 3422 {
3420 struct list_head *cpu_list; 3423 struct list_head *cpu_list;
3421 unsigned long flags; 3424 unsigned long flags;
3422 3425
3423 BUG_ON(!req->q->softirq_done_fn); 3426 BUG_ON(!req->q->softirq_done_fn);
3424 3427
3425 local_irq_save(flags); 3428 local_irq_save(flags);
3426 3429
3427 cpu_list = &__get_cpu_var(blk_cpu_done); 3430 cpu_list = &__get_cpu_var(blk_cpu_done);
3428 list_add_tail(&req->donelist, cpu_list); 3431 list_add_tail(&req->donelist, cpu_list);
3429 raise_softirq_irqoff(BLOCK_SOFTIRQ); 3432 raise_softirq_irqoff(BLOCK_SOFTIRQ);
3430 3433
3431 local_irq_restore(flags); 3434 local_irq_restore(flags);
3432 } 3435 }
3433 3436
3434 EXPORT_SYMBOL(blk_complete_request); 3437 EXPORT_SYMBOL(blk_complete_request);
3435 3438
3436 /* 3439 /*
3437 * queue lock must be held 3440 * queue lock must be held
3438 */ 3441 */
3439 void end_that_request_last(struct request *req, int uptodate) 3442 void end_that_request_last(struct request *req, int uptodate)
3440 { 3443 {
3441 struct gendisk *disk = req->rq_disk; 3444 struct gendisk *disk = req->rq_disk;
3442 int error; 3445 int error;
3443 3446
3444 /* 3447 /*
3445 * extend uptodate bool to allow < 0 value to be direct io error 3448 * extend uptodate bool to allow < 0 value to be direct io error
3446 */ 3449 */
3447 error = 0; 3450 error = 0;
3448 if (end_io_error(uptodate)) 3451 if (end_io_error(uptodate))
3449 error = !uptodate ? -EIO : uptodate; 3452 error = !uptodate ? -EIO : uptodate;
3450 3453
3451 if (unlikely(laptop_mode) && blk_fs_request(req)) 3454 if (unlikely(laptop_mode) && blk_fs_request(req))
3452 laptop_io_completion(); 3455 laptop_io_completion();
3453 3456
3454 /* 3457 /*
3455 * Account IO completion. bar_rq isn't accounted as a normal 3458 * Account IO completion. bar_rq isn't accounted as a normal
3456 * IO on queueing nor completion. Accounting the containing 3459 * IO on queueing nor completion. Accounting the containing
3457 * request is enough. 3460 * request is enough.
3458 */ 3461 */
3459 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) { 3462 if (disk && blk_fs_request(req) && req != &req->q->bar_rq) {
3460 unsigned long duration = jiffies - req->start_time; 3463 unsigned long duration = jiffies - req->start_time;
3461 const int rw = rq_data_dir(req); 3464 const int rw = rq_data_dir(req);
3462 3465
3463 __disk_stat_inc(disk, ios[rw]); 3466 __disk_stat_inc(disk, ios[rw]);
3464 __disk_stat_add(disk, ticks[rw], duration); 3467 __disk_stat_add(disk, ticks[rw], duration);
3465 disk_round_stats(disk); 3468 disk_round_stats(disk);
3466 disk->in_flight--; 3469 disk->in_flight--;
3467 } 3470 }
3468 if (req->end_io) 3471 if (req->end_io)
3469 req->end_io(req, error); 3472 req->end_io(req, error);
3470 else 3473 else
3471 __blk_put_request(req->q, req); 3474 __blk_put_request(req->q, req);
3472 } 3475 }
3473 3476
3474 EXPORT_SYMBOL(end_that_request_last); 3477 EXPORT_SYMBOL(end_that_request_last);
3475 3478
3476 void end_request(struct request *req, int uptodate) 3479 void end_request(struct request *req, int uptodate)
3477 { 3480 {
3478 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) { 3481 if (!end_that_request_first(req, uptodate, req->hard_cur_sectors)) {
3479 add_disk_randomness(req->rq_disk); 3482 add_disk_randomness(req->rq_disk);
3480 blkdev_dequeue_request(req); 3483 blkdev_dequeue_request(req);
3481 end_that_request_last(req, uptodate); 3484 end_that_request_last(req, uptodate);
3482 } 3485 }
3483 } 3486 }
3484 3487
3485 EXPORT_SYMBOL(end_request); 3488 EXPORT_SYMBOL(end_request);
3486 3489
3487 void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio) 3490 void blk_rq_bio_prep(request_queue_t *q, struct request *rq, struct bio *bio)
3488 { 3491 {
3489 /* first three bits are identical in rq->flags and bio->bi_rw */ 3492 /* first three bits are identical in rq->flags and bio->bi_rw */
3490 rq->flags |= (bio->bi_rw & 7); 3493 rq->flags |= (bio->bi_rw & 7);
3491 3494
3492 rq->nr_phys_segments = bio_phys_segments(q, bio); 3495 rq->nr_phys_segments = bio_phys_segments(q, bio);
3493 rq->nr_hw_segments = bio_hw_segments(q, bio); 3496 rq->nr_hw_segments = bio_hw_segments(q, bio);
3494 rq->current_nr_sectors = bio_cur_sectors(bio); 3497 rq->current_nr_sectors = bio_cur_sectors(bio);
3495 rq->hard_cur_sectors = rq->current_nr_sectors; 3498 rq->hard_cur_sectors = rq->current_nr_sectors;
3496 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio); 3499 rq->hard_nr_sectors = rq->nr_sectors = bio_sectors(bio);
3497 rq->buffer = bio_data(bio); 3500 rq->buffer = bio_data(bio);
3498 3501
3499 rq->bio = rq->biotail = bio; 3502 rq->bio = rq->biotail = bio;
3500 } 3503 }
3501 3504
3502 EXPORT_SYMBOL(blk_rq_bio_prep); 3505 EXPORT_SYMBOL(blk_rq_bio_prep);
3503 3506
3504 int kblockd_schedule_work(struct work_struct *work) 3507 int kblockd_schedule_work(struct work_struct *work)
3505 { 3508 {
3506 return queue_work(kblockd_workqueue, work); 3509 return queue_work(kblockd_workqueue, work);
3507 } 3510 }
3508 3511
3509 EXPORT_SYMBOL(kblockd_schedule_work); 3512 EXPORT_SYMBOL(kblockd_schedule_work);
3510 3513
3511 void kblockd_flush(void) 3514 void kblockd_flush(void)
3512 { 3515 {
3513 flush_workqueue(kblockd_workqueue); 3516 flush_workqueue(kblockd_workqueue);
3514 } 3517 }
3515 EXPORT_SYMBOL(kblockd_flush); 3518 EXPORT_SYMBOL(kblockd_flush);
3516 3519
3517 int __init blk_dev_init(void) 3520 int __init blk_dev_init(void)
3518 { 3521 {
3519 int i; 3522 int i;
3520 3523
3521 kblockd_workqueue = create_workqueue("kblockd"); 3524 kblockd_workqueue = create_workqueue("kblockd");
3522 if (!kblockd_workqueue) 3525 if (!kblockd_workqueue)
3523 panic("Failed to create kblockd\n"); 3526 panic("Failed to create kblockd\n");
3524 3527
3525 request_cachep = kmem_cache_create("blkdev_requests", 3528 request_cachep = kmem_cache_create("blkdev_requests",
3526 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL); 3529 sizeof(struct request), 0, SLAB_PANIC, NULL, NULL);
3527 3530
3528 requestq_cachep = kmem_cache_create("blkdev_queue", 3531 requestq_cachep = kmem_cache_create("blkdev_queue",
3529 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL); 3532 sizeof(request_queue_t), 0, SLAB_PANIC, NULL, NULL);
3530 3533
3531 iocontext_cachep = kmem_cache_create("blkdev_ioc", 3534 iocontext_cachep = kmem_cache_create("blkdev_ioc",
3532 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL); 3535 sizeof(struct io_context), 0, SLAB_PANIC, NULL, NULL);
3533 3536
3534 for_each_possible_cpu(i) 3537 for_each_possible_cpu(i)
3535 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i)); 3538 INIT_LIST_HEAD(&per_cpu(blk_cpu_done, i));
3536 3539
3537 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL); 3540 open_softirq(BLOCK_SOFTIRQ, blk_done_softirq, NULL);
3538 #ifdef CONFIG_HOTPLUG_CPU 3541 #ifdef CONFIG_HOTPLUG_CPU
3539 register_cpu_notifier(&blk_cpu_notifier); 3542 register_cpu_notifier(&blk_cpu_notifier);
3540 #endif 3543 #endif
3541 3544
3542 blk_max_low_pfn = max_low_pfn; 3545 blk_max_low_pfn = max_low_pfn;
3543 blk_max_pfn = max_pfn; 3546 blk_max_pfn = max_pfn;
3544 3547
3545 return 0; 3548 return 0;
3546 } 3549 }
3547 3550
3548 /* 3551 /*
3549 * IO Context helper functions 3552 * IO Context helper functions
3550 */ 3553 */
3551 void put_io_context(struct io_context *ioc) 3554 void put_io_context(struct io_context *ioc)
3552 { 3555 {
3553 if (ioc == NULL) 3556 if (ioc == NULL)
3554 return; 3557 return;
3555 3558
3556 BUG_ON(atomic_read(&ioc->refcount) == 0); 3559 BUG_ON(atomic_read(&ioc->refcount) == 0);
3557 3560
3558 if (atomic_dec_and_test(&ioc->refcount)) { 3561 if (atomic_dec_and_test(&ioc->refcount)) {
3559 struct cfq_io_context *cic; 3562 struct cfq_io_context *cic;
3560 3563
3561 rcu_read_lock(); 3564 rcu_read_lock();
3562 if (ioc->aic && ioc->aic->dtor) 3565 if (ioc->aic && ioc->aic->dtor)
3563 ioc->aic->dtor(ioc->aic); 3566 ioc->aic->dtor(ioc->aic);
3564 if (ioc->cic_root.rb_node != NULL) { 3567 if (ioc->cic_root.rb_node != NULL) {
3565 struct rb_node *n = rb_first(&ioc->cic_root); 3568 struct rb_node *n = rb_first(&ioc->cic_root);
3566 3569
3567 cic = rb_entry(n, struct cfq_io_context, rb_node); 3570 cic = rb_entry(n, struct cfq_io_context, rb_node);
3568 cic->dtor(ioc); 3571 cic->dtor(ioc);
3569 } 3572 }
3570 rcu_read_unlock(); 3573 rcu_read_unlock();
3571 3574
3572 kmem_cache_free(iocontext_cachep, ioc); 3575 kmem_cache_free(iocontext_cachep, ioc);
3573 } 3576 }
3574 } 3577 }
3575 EXPORT_SYMBOL(put_io_context); 3578 EXPORT_SYMBOL(put_io_context);
3576 3579
3577 /* Called by the exitting task */ 3580 /* Called by the exitting task */
3578 void exit_io_context(void) 3581 void exit_io_context(void)
3579 { 3582 {
3580 unsigned long flags; 3583 unsigned long flags;
3581 struct io_context *ioc; 3584 struct io_context *ioc;
3582 struct cfq_io_context *cic; 3585 struct cfq_io_context *cic;
3583 3586
3584 local_irq_save(flags); 3587 local_irq_save(flags);
3585 task_lock(current); 3588 task_lock(current);
3586 ioc = current->io_context; 3589 ioc = current->io_context;
3587 current->io_context = NULL; 3590 current->io_context = NULL;
3588 ioc->task = NULL; 3591 ioc->task = NULL;
3589 task_unlock(current); 3592 task_unlock(current);
3590 local_irq_restore(flags); 3593 local_irq_restore(flags);
3591 3594
3592 if (ioc->aic && ioc->aic->exit) 3595 if (ioc->aic && ioc->aic->exit)
3593 ioc->aic->exit(ioc->aic); 3596 ioc->aic->exit(ioc->aic);
3594 if (ioc->cic_root.rb_node != NULL) { 3597 if (ioc->cic_root.rb_node != NULL) {
3595 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node); 3598 cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
3596 cic->exit(ioc); 3599 cic->exit(ioc);
3597 } 3600 }
3598 3601
3599 put_io_context(ioc); 3602 put_io_context(ioc);
3600 } 3603 }
3601 3604
3602 /* 3605 /*
3603 * If the current task has no IO context then create one and initialise it. 3606 * If the current task has no IO context then create one and initialise it.
3604 * Otherwise, return its existing IO context. 3607 * Otherwise, return its existing IO context.
3605 * 3608 *
3606 * This returned IO context doesn't have a specifically elevated refcount, 3609 * This returned IO context doesn't have a specifically elevated refcount,
3607 * but since the current task itself holds a reference, the context can be 3610 * but since the current task itself holds a reference, the context can be
3608 * used in general code, so long as it stays within `current` context. 3611 * used in general code, so long as it stays within `current` context.
3609 */ 3612 */
3610 struct io_context *current_io_context(gfp_t gfp_flags) 3613 struct io_context *current_io_context(gfp_t gfp_flags)
3611 { 3614 {
3612 struct task_struct *tsk = current; 3615 struct task_struct *tsk = current;
3613 struct io_context *ret; 3616 struct io_context *ret;
3614 3617
3615 ret = tsk->io_context; 3618 ret = tsk->io_context;
3616 if (likely(ret)) 3619 if (likely(ret))
3617 return ret; 3620 return ret;
3618 3621
3619 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags); 3622 ret = kmem_cache_alloc(iocontext_cachep, gfp_flags);
3620 if (ret) { 3623 if (ret) {
3621 atomic_set(&ret->refcount, 1); 3624 atomic_set(&ret->refcount, 1);
3622 ret->task = current; 3625 ret->task = current;
3623 ret->set_ioprio = NULL; 3626 ret->set_ioprio = NULL;
3624 ret->last_waited = jiffies; /* doesn't matter... */ 3627 ret->last_waited = jiffies; /* doesn't matter... */
3625 ret->nr_batch_requests = 0; /* because this is 0 */ 3628 ret->nr_batch_requests = 0; /* because this is 0 */
3626 ret->aic = NULL; 3629 ret->aic = NULL;
3627 ret->cic_root.rb_node = NULL; 3630 ret->cic_root.rb_node = NULL;
3628 tsk->io_context = ret; 3631 tsk->io_context = ret;
3629 } 3632 }
3630 3633
3631 return ret; 3634 return ret;
3632 } 3635 }
3633 EXPORT_SYMBOL(current_io_context); 3636 EXPORT_SYMBOL(current_io_context);
3634 3637
3635 /* 3638 /*
3636 * If the current task has no IO context then create one and initialise it. 3639 * If the current task has no IO context then create one and initialise it.
3637 * If it does have a context, take a ref on it. 3640 * If it does have a context, take a ref on it.
3638 * 3641 *
3639 * This is always called in the context of the task which submitted the I/O. 3642 * This is always called in the context of the task which submitted the I/O.
3640 */ 3643 */
3641 struct io_context *get_io_context(gfp_t gfp_flags) 3644 struct io_context *get_io_context(gfp_t gfp_flags)
3642 { 3645 {
3643 struct io_context *ret; 3646 struct io_context *ret;
3644 ret = current_io_context(gfp_flags); 3647 ret = current_io_context(gfp_flags);
3645 if (likely(ret)) 3648 if (likely(ret))
3646 atomic_inc(&ret->refcount); 3649 atomic_inc(&ret->refcount);
3647 return ret; 3650 return ret;
3648 } 3651 }
3649 EXPORT_SYMBOL(get_io_context); 3652 EXPORT_SYMBOL(get_io_context);
3650 3653
3651 void copy_io_context(struct io_context **pdst, struct io_context **psrc) 3654 void copy_io_context(struct io_context **pdst, struct io_context **psrc)
3652 { 3655 {
3653 struct io_context *src = *psrc; 3656 struct io_context *src = *psrc;
3654 struct io_context *dst = *pdst; 3657 struct io_context *dst = *pdst;
3655 3658
3656 if (src) { 3659 if (src) {
3657 BUG_ON(atomic_read(&src->refcount) == 0); 3660 BUG_ON(atomic_read(&src->refcount) == 0);
3658 atomic_inc(&src->refcount); 3661 atomic_inc(&src->refcount);
3659 put_io_context(dst); 3662 put_io_context(dst);
3660 *pdst = src; 3663 *pdst = src;
3661 } 3664 }
3662 } 3665 }
3663 EXPORT_SYMBOL(copy_io_context); 3666 EXPORT_SYMBOL(copy_io_context);
3664 3667
3665 void swap_io_context(struct io_context **ioc1, struct io_context **ioc2) 3668 void swap_io_context(struct io_context **ioc1, struct io_context **ioc2)
3666 { 3669 {
3667 struct io_context *temp; 3670 struct io_context *temp;
3668 temp = *ioc1; 3671 temp = *ioc1;
3669 *ioc1 = *ioc2; 3672 *ioc1 = *ioc2;
3670 *ioc2 = temp; 3673 *ioc2 = temp;
3671 } 3674 }
3672 EXPORT_SYMBOL(swap_io_context); 3675 EXPORT_SYMBOL(swap_io_context);
3673 3676
3674 /* 3677 /*
3675 * sysfs parts below 3678 * sysfs parts below
3676 */ 3679 */
3677 struct queue_sysfs_entry { 3680 struct queue_sysfs_entry {
3678 struct attribute attr; 3681 struct attribute attr;
3679 ssize_t (*show)(struct request_queue *, char *); 3682 ssize_t (*show)(struct request_queue *, char *);
3680 ssize_t (*store)(struct request_queue *, const char *, size_t); 3683 ssize_t (*store)(struct request_queue *, const char *, size_t);
3681 }; 3684 };
3682 3685
3683 static ssize_t 3686 static ssize_t
3684 queue_var_show(unsigned int var, char *page) 3687 queue_var_show(unsigned int var, char *page)
3685 { 3688 {
3686 return sprintf(page, "%d\n", var); 3689 return sprintf(page, "%d\n", var);
3687 } 3690 }
3688 3691
3689 static ssize_t 3692 static ssize_t
3690 queue_var_store(unsigned long *var, const char *page, size_t count) 3693 queue_var_store(unsigned long *var, const char *page, size_t count)
3691 { 3694 {
3692 char *p = (char *) page; 3695 char *p = (char *) page;
3693 3696
3694 *var = simple_strtoul(p, &p, 10); 3697 *var = simple_strtoul(p, &p, 10);
3695 return count; 3698 return count;
3696 } 3699 }
3697 3700
3698 static ssize_t queue_requests_show(struct request_queue *q, char *page) 3701 static ssize_t queue_requests_show(struct request_queue *q, char *page)
3699 { 3702 {
3700 return queue_var_show(q->nr_requests, (page)); 3703 return queue_var_show(q->nr_requests, (page));
3701 } 3704 }
3702 3705
3703 static ssize_t 3706 static ssize_t
3704 queue_requests_store(struct request_queue *q, const char *page, size_t count) 3707 queue_requests_store(struct request_queue *q, const char *page, size_t count)
3705 { 3708 {
3706 struct request_list *rl = &q->rq; 3709 struct request_list *rl = &q->rq;
3707 unsigned long nr; 3710 unsigned long nr;
3708 int ret = queue_var_store(&nr, page, count); 3711 int ret = queue_var_store(&nr, page, count);
3709 if (nr < BLKDEV_MIN_RQ) 3712 if (nr < BLKDEV_MIN_RQ)
3710 nr = BLKDEV_MIN_RQ; 3713 nr = BLKDEV_MIN_RQ;
3711 3714
3712 spin_lock_irq(q->queue_lock); 3715 spin_lock_irq(q->queue_lock);
3713 q->nr_requests = nr; 3716 q->nr_requests = nr;
3714 blk_queue_congestion_threshold(q); 3717 blk_queue_congestion_threshold(q);
3715 3718
3716 if (rl->count[READ] >= queue_congestion_on_threshold(q)) 3719 if (rl->count[READ] >= queue_congestion_on_threshold(q))
3717 set_queue_congested(q, READ); 3720 set_queue_congested(q, READ);
3718 else if (rl->count[READ] < queue_congestion_off_threshold(q)) 3721 else if (rl->count[READ] < queue_congestion_off_threshold(q))
3719 clear_queue_congested(q, READ); 3722 clear_queue_congested(q, READ);
3720 3723
3721 if (rl->count[WRITE] >= queue_congestion_on_threshold(q)) 3724 if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
3722 set_queue_congested(q, WRITE); 3725 set_queue_congested(q, WRITE);
3723 else if (rl->count[WRITE] < queue_congestion_off_threshold(q)) 3726 else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
3724 clear_queue_congested(q, WRITE); 3727 clear_queue_congested(q, WRITE);
3725 3728
3726 if (rl->count[READ] >= q->nr_requests) { 3729 if (rl->count[READ] >= q->nr_requests) {
3727 blk_set_queue_full(q, READ); 3730 blk_set_queue_full(q, READ);
3728 } else if (rl->count[READ]+1 <= q->nr_requests) { 3731 } else if (rl->count[READ]+1 <= q->nr_requests) {
3729 blk_clear_queue_full(q, READ); 3732 blk_clear_queue_full(q, READ);
3730 wake_up(&rl->wait[READ]); 3733 wake_up(&rl->wait[READ]);
3731 } 3734 }
3732 3735
3733 if (rl->count[WRITE] >= q->nr_requests) { 3736 if (rl->count[WRITE] >= q->nr_requests) {
3734 blk_set_queue_full(q, WRITE); 3737 blk_set_queue_full(q, WRITE);
3735 } else if (rl->count[WRITE]+1 <= q->nr_requests) { 3738 } else if (rl->count[WRITE]+1 <= q->nr_requests) {
3736 blk_clear_queue_full(q, WRITE); 3739 blk_clear_queue_full(q, WRITE);
3737 wake_up(&rl->wait[WRITE]); 3740 wake_up(&rl->wait[WRITE]);
3738 } 3741 }
3739 spin_unlock_irq(q->queue_lock); 3742 spin_unlock_irq(q->queue_lock);
3740 return ret; 3743 return ret;
3741 } 3744 }
3742 3745
3743 static ssize_t queue_ra_show(struct request_queue *q, char *page) 3746 static ssize_t queue_ra_show(struct request_queue *q, char *page)
3744 { 3747 {
3745 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); 3748 int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3746 3749
3747 return queue_var_show(ra_kb, (page)); 3750 return queue_var_show(ra_kb, (page));
3748 } 3751 }
3749 3752
3750 static ssize_t 3753 static ssize_t
3751 queue_ra_store(struct request_queue *q, const char *page, size_t count) 3754 queue_ra_store(struct request_queue *q, const char *page, size_t count)
3752 { 3755 {
3753 unsigned long ra_kb; 3756 unsigned long ra_kb;
3754 ssize_t ret = queue_var_store(&ra_kb, page, count); 3757 ssize_t ret = queue_var_store(&ra_kb, page, count);
3755 3758
3756 spin_lock_irq(q->queue_lock); 3759 spin_lock_irq(q->queue_lock);
3757 if (ra_kb > (q->max_sectors >> 1)) 3760 if (ra_kb > (q->max_sectors >> 1))
3758 ra_kb = (q->max_sectors >> 1); 3761 ra_kb = (q->max_sectors >> 1);
3759 3762
3760 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); 3763 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
3761 spin_unlock_irq(q->queue_lock); 3764 spin_unlock_irq(q->queue_lock);
3762 3765
3763 return ret; 3766 return ret;
3764 } 3767 }
3765 3768
3766 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) 3769 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
3767 { 3770 {
3768 int max_sectors_kb = q->max_sectors >> 1; 3771 int max_sectors_kb = q->max_sectors >> 1;
3769 3772
3770 return queue_var_show(max_sectors_kb, (page)); 3773 return queue_var_show(max_sectors_kb, (page));
3771 } 3774 }
3772 3775
3773 static ssize_t 3776 static ssize_t
3774 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) 3777 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
3775 { 3778 {
3776 unsigned long max_sectors_kb, 3779 unsigned long max_sectors_kb,
3777 max_hw_sectors_kb = q->max_hw_sectors >> 1, 3780 max_hw_sectors_kb = q->max_hw_sectors >> 1,
3778 page_kb = 1 << (PAGE_CACHE_SHIFT - 10); 3781 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
3779 ssize_t ret = queue_var_store(&max_sectors_kb, page, count); 3782 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
3780 int ra_kb; 3783 int ra_kb;
3781 3784
3782 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) 3785 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
3783 return -EINVAL; 3786 return -EINVAL;
3784 /* 3787 /*
3785 * Take the queue lock to update the readahead and max_sectors 3788 * Take the queue lock to update the readahead and max_sectors
3786 * values synchronously: 3789 * values synchronously:
3787 */ 3790 */
3788 spin_lock_irq(q->queue_lock); 3791 spin_lock_irq(q->queue_lock);
3789 /* 3792 /*
3790 * Trim readahead window as well, if necessary: 3793 * Trim readahead window as well, if necessary:
3791 */ 3794 */
3792 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10); 3795 ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
3793 if (ra_kb > max_sectors_kb) 3796 if (ra_kb > max_sectors_kb)
3794 q->backing_dev_info.ra_pages = 3797 q->backing_dev_info.ra_pages =
3795 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10); 3798 max_sectors_kb >> (PAGE_CACHE_SHIFT - 10);
3796 3799
3797 q->max_sectors = max_sectors_kb << 1; 3800 q->max_sectors = max_sectors_kb << 1;
3798 spin_unlock_irq(q->queue_lock); 3801 spin_unlock_irq(q->queue_lock);
3799 3802
3800 return ret; 3803 return ret;
3801 } 3804 }
3802 3805
3803 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) 3806 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
3804 { 3807 {
3805 int max_hw_sectors_kb = q->max_hw_sectors >> 1; 3808 int max_hw_sectors_kb = q->max_hw_sectors >> 1;
3806 3809
3807 return queue_var_show(max_hw_sectors_kb, (page)); 3810 return queue_var_show(max_hw_sectors_kb, (page));
3808 } 3811 }
3809 3812
3810 3813
3811 static struct queue_sysfs_entry queue_requests_entry = { 3814 static struct queue_sysfs_entry queue_requests_entry = {
3812 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, 3815 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
3813 .show = queue_requests_show, 3816 .show = queue_requests_show,
3814 .store = queue_requests_store, 3817 .store = queue_requests_store,
3815 }; 3818 };
3816 3819
3817 static struct queue_sysfs_entry queue_ra_entry = { 3820 static struct queue_sysfs_entry queue_ra_entry = {
3818 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, 3821 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
3819 .show = queue_ra_show, 3822 .show = queue_ra_show,
3820 .store = queue_ra_store, 3823 .store = queue_ra_store,
3821 }; 3824 };
3822 3825
3823 static struct queue_sysfs_entry queue_max_sectors_entry = { 3826 static struct queue_sysfs_entry queue_max_sectors_entry = {
3824 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, 3827 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
3825 .show = queue_max_sectors_show, 3828 .show = queue_max_sectors_show,
3826 .store = queue_max_sectors_store, 3829 .store = queue_max_sectors_store,
3827 }; 3830 };
3828 3831
3829 static struct queue_sysfs_entry queue_max_hw_sectors_entry = { 3832 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
3830 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, 3833 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
3831 .show = queue_max_hw_sectors_show, 3834 .show = queue_max_hw_sectors_show,
3832 }; 3835 };
3833 3836
3834 static struct queue_sysfs_entry queue_iosched_entry = { 3837 static struct queue_sysfs_entry queue_iosched_entry = {
3835 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, 3838 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
3836 .show = elv_iosched_show, 3839 .show = elv_iosched_show,
3837 .store = elv_iosched_store, 3840 .store = elv_iosched_store,
3838 }; 3841 };
3839 3842
3840 static struct attribute *default_attrs[] = { 3843 static struct attribute *default_attrs[] = {
3841 &queue_requests_entry.attr, 3844 &queue_requests_entry.attr,
3842 &queue_ra_entry.attr, 3845 &queue_ra_entry.attr,
3843 &queue_max_hw_sectors_entry.attr, 3846 &queue_max_hw_sectors_entry.attr,
3844 &queue_max_sectors_entry.attr, 3847 &queue_max_sectors_entry.attr,
3845 &queue_iosched_entry.attr, 3848 &queue_iosched_entry.attr,
3846 NULL, 3849 NULL,
3847 }; 3850 };
3848 3851
3849 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) 3852 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
3850 3853
3851 static ssize_t 3854 static ssize_t
3852 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 3855 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3853 { 3856 {
3854 struct queue_sysfs_entry *entry = to_queue(attr); 3857 struct queue_sysfs_entry *entry = to_queue(attr);
3855 request_queue_t *q = container_of(kobj, struct request_queue, kobj); 3858 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3856 ssize_t res; 3859 ssize_t res;
3857 3860
3858 if (!entry->show) 3861 if (!entry->show)
3859 return -EIO; 3862 return -EIO;
3860 mutex_lock(&q->sysfs_lock); 3863 mutex_lock(&q->sysfs_lock);
3861 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { 3864 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3862 mutex_unlock(&q->sysfs_lock); 3865 mutex_unlock(&q->sysfs_lock);
3863 return -ENOENT; 3866 return -ENOENT;
3864 } 3867 }
3865 res = entry->show(q, page); 3868 res = entry->show(q, page);
3866 mutex_unlock(&q->sysfs_lock); 3869 mutex_unlock(&q->sysfs_lock);
3867 return res; 3870 return res;
3868 } 3871 }
3869 3872
3870 static ssize_t 3873 static ssize_t
3871 queue_attr_store(struct kobject *kobj, struct attribute *attr, 3874 queue_attr_store(struct kobject *kobj, struct attribute *attr,
3872 const char *page, size_t length) 3875 const char *page, size_t length)
3873 { 3876 {
3874 struct queue_sysfs_entry *entry = to_queue(attr); 3877 struct queue_sysfs_entry *entry = to_queue(attr);
3875 request_queue_t *q = container_of(kobj, struct request_queue, kobj); 3878 request_queue_t *q = container_of(kobj, struct request_queue, kobj);
3876 3879
3877 ssize_t res; 3880 ssize_t res;
3878 3881
3879 if (!entry->store) 3882 if (!entry->store)
3880 return -EIO; 3883 return -EIO;
3881 mutex_lock(&q->sysfs_lock); 3884 mutex_lock(&q->sysfs_lock);
3882 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) { 3885 if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
3883 mutex_unlock(&q->sysfs_lock); 3886 mutex_unlock(&q->sysfs_lock);
3884 return -ENOENT; 3887 return -ENOENT;
3885 } 3888 }
3886 res = entry->store(q, page, length); 3889 res = entry->store(q, page, length);
3887 mutex_unlock(&q->sysfs_lock); 3890 mutex_unlock(&q->sysfs_lock);
3888 return res; 3891 return res;
3889 } 3892 }
3890 3893
3891 static struct sysfs_ops queue_sysfs_ops = { 3894 static struct sysfs_ops queue_sysfs_ops = {
3892 .show = queue_attr_show, 3895 .show = queue_attr_show,
3893 .store = queue_attr_store, 3896 .store = queue_attr_store,
3894 }; 3897 };
3895 3898
3896 static struct kobj_type queue_ktype = { 3899 static struct kobj_type queue_ktype = {
3897 .sysfs_ops = &queue_sysfs_ops, 3900 .sysfs_ops = &queue_sysfs_ops,
3898 .default_attrs = default_attrs, 3901 .default_attrs = default_attrs,
3899 .release = blk_release_queue, 3902 .release = blk_release_queue,
3900 }; 3903 };
3901 3904
3902 int blk_register_queue(struct gendisk *disk) 3905 int blk_register_queue(struct gendisk *disk)
3903 { 3906 {
3904 int ret; 3907 int ret;
3905 3908
3906 request_queue_t *q = disk->queue; 3909 request_queue_t *q = disk->queue;
3907 3910
3908 if (!q || !q->request_fn) 3911 if (!q || !q->request_fn)
3909 return -ENXIO; 3912 return -ENXIO;
3910 3913
3911 q->kobj.parent = kobject_get(&disk->kobj); 3914 q->kobj.parent = kobject_get(&disk->kobj);
3912 3915
3913 ret = kobject_add(&q->kobj); 3916 ret = kobject_add(&q->kobj);
3914 if (ret < 0) 3917 if (ret < 0)
3915 return ret; 3918 return ret;
3916 3919
3917 kobject_uevent(&q->kobj, KOBJ_ADD); 3920 kobject_uevent(&q->kobj, KOBJ_ADD);
3918 3921
3919 ret = elv_register_queue(q); 3922 ret = elv_register_queue(q);
3920 if (ret) { 3923 if (ret) {
3921 kobject_uevent(&q->kobj, KOBJ_REMOVE); 3924 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3922 kobject_del(&q->kobj); 3925 kobject_del(&q->kobj);
3923 return ret; 3926 return ret;
3924 } 3927 }
3925 3928
3926 return 0; 3929 return 0;
3927 } 3930 }
3928 3931
3929 void blk_unregister_queue(struct gendisk *disk) 3932 void blk_unregister_queue(struct gendisk *disk)
3930 { 3933 {
3931 request_queue_t *q = disk->queue; 3934 request_queue_t *q = disk->queue;
3932 3935
3933 if (q && q->request_fn) { 3936 if (q && q->request_fn) {
3934 elv_unregister_queue(q); 3937 elv_unregister_queue(q);
3935 3938
3936 kobject_uevent(&q->kobj, KOBJ_REMOVE); 3939 kobject_uevent(&q->kobj, KOBJ_REMOVE);
3937 kobject_del(&q->kobj); 3940 kobject_del(&q->kobj);
3938 kobject_put(&disk->kobj); 3941 kobject_put(&disk->kobj);
3939 } 3942 }
3940 } 3943 }
3941 3944