block: add support for IO CPU affinity
[linux-2.6.git] / block / blk-sysfs.c
1 /*
2  * Functions related to sysfs handling
3  */
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/bio.h>
7 #include <linux/blkdev.h>
8 #include <linux/blktrace_api.h>
9
10 #include "blk.h"
11
12 struct queue_sysfs_entry {
13         struct attribute attr;
14         ssize_t (*show)(struct request_queue *, char *);
15         ssize_t (*store)(struct request_queue *, const char *, size_t);
16 };
17
18 static ssize_t
19 queue_var_show(unsigned int var, char *page)
20 {
21         return sprintf(page, "%d\n", var);
22 }
23
24 static ssize_t
25 queue_var_store(unsigned long *var, const char *page, size_t count)
26 {
27         char *p = (char *) page;
28
29         *var = simple_strtoul(p, &p, 10);
30         return count;
31 }
32
33 static ssize_t queue_requests_show(struct request_queue *q, char *page)
34 {
35         return queue_var_show(q->nr_requests, (page));
36 }
37
38 static ssize_t
39 queue_requests_store(struct request_queue *q, const char *page, size_t count)
40 {
41         struct request_list *rl = &q->rq;
42         unsigned long nr;
43         int ret = queue_var_store(&nr, page, count);
44         if (nr < BLKDEV_MIN_RQ)
45                 nr = BLKDEV_MIN_RQ;
46
47         spin_lock_irq(q->queue_lock);
48         q->nr_requests = nr;
49         blk_queue_congestion_threshold(q);
50
51         if (rl->count[READ] >= queue_congestion_on_threshold(q))
52                 blk_set_queue_congested(q, READ);
53         else if (rl->count[READ] < queue_congestion_off_threshold(q))
54                 blk_clear_queue_congested(q, READ);
55
56         if (rl->count[WRITE] >= queue_congestion_on_threshold(q))
57                 blk_set_queue_congested(q, WRITE);
58         else if (rl->count[WRITE] < queue_congestion_off_threshold(q))
59                 blk_clear_queue_congested(q, WRITE);
60
61         if (rl->count[READ] >= q->nr_requests) {
62                 blk_set_queue_full(q, READ);
63         } else if (rl->count[READ]+1 <= q->nr_requests) {
64                 blk_clear_queue_full(q, READ);
65                 wake_up(&rl->wait[READ]);
66         }
67
68         if (rl->count[WRITE] >= q->nr_requests) {
69                 blk_set_queue_full(q, WRITE);
70         } else if (rl->count[WRITE]+1 <= q->nr_requests) {
71                 blk_clear_queue_full(q, WRITE);
72                 wake_up(&rl->wait[WRITE]);
73         }
74         spin_unlock_irq(q->queue_lock);
75         return ret;
76 }
77
78 static ssize_t queue_ra_show(struct request_queue *q, char *page)
79 {
80         int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
81
82         return queue_var_show(ra_kb, (page));
83 }
84
85 static ssize_t
86 queue_ra_store(struct request_queue *q, const char *page, size_t count)
87 {
88         unsigned long ra_kb;
89         ssize_t ret = queue_var_store(&ra_kb, page, count);
90
91         spin_lock_irq(q->queue_lock);
92         q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
93         spin_unlock_irq(q->queue_lock);
94
95         return ret;
96 }
97
98 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
99 {
100         int max_sectors_kb = q->max_sectors >> 1;
101
102         return queue_var_show(max_sectors_kb, (page));
103 }
104
105 static ssize_t queue_hw_sector_size_show(struct request_queue *q, char *page)
106 {
107         return queue_var_show(q->hardsect_size, page);
108 }
109
110 static ssize_t
111 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
112 {
113         unsigned long max_sectors_kb,
114                         max_hw_sectors_kb = q->max_hw_sectors >> 1,
115                         page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
116         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
117
118         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
119                 return -EINVAL;
120         /*
121          * Take the queue lock to update the readahead and max_sectors
122          * values synchronously:
123          */
124         spin_lock_irq(q->queue_lock);
125         q->max_sectors = max_sectors_kb << 1;
126         spin_unlock_irq(q->queue_lock);
127
128         return ret;
129 }
130
131 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
132 {
133         int max_hw_sectors_kb = q->max_hw_sectors >> 1;
134
135         return queue_var_show(max_hw_sectors_kb, (page));
136 }
137
138 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
139 {
140         return queue_var_show(blk_queue_nomerges(q), page);
141 }
142
143 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
144                                     size_t count)
145 {
146         unsigned long nm;
147         ssize_t ret = queue_var_store(&nm, page, count);
148
149         spin_lock_irq(q->queue_lock);
150         if (nm)
151                 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
152         else
153                 queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
154
155         spin_unlock_irq(q->queue_lock);
156         return ret;
157 }
158
159 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
160 {
161         unsigned int set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
162
163         return queue_var_show(set != 0, page);
164 }
165
166 static ssize_t
167 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
168 {
169         ssize_t ret = -EINVAL;
170 #if defined(CONFIG_USE_GENERIC_SMP_HELPERS)
171         unsigned long val;
172
173         ret = queue_var_store(&val, page, count);
174         spin_lock_irq(q->queue_lock);
175         if (val)
176                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
177         else
178                 queue_flag_clear(QUEUE_FLAG_SAME_COMP,  q);
179         spin_unlock_irq(q->queue_lock);
180 #endif
181         return ret;
182 }
183
184 static struct queue_sysfs_entry queue_requests_entry = {
185         .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
186         .show = queue_requests_show,
187         .store = queue_requests_store,
188 };
189
190 static struct queue_sysfs_entry queue_ra_entry = {
191         .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
192         .show = queue_ra_show,
193         .store = queue_ra_store,
194 };
195
196 static struct queue_sysfs_entry queue_max_sectors_entry = {
197         .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
198         .show = queue_max_sectors_show,
199         .store = queue_max_sectors_store,
200 };
201
202 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
203         .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
204         .show = queue_max_hw_sectors_show,
205 };
206
207 static struct queue_sysfs_entry queue_iosched_entry = {
208         .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
209         .show = elv_iosched_show,
210         .store = elv_iosched_store,
211 };
212
213 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
214         .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
215         .show = queue_hw_sector_size_show,
216 };
217
218 static struct queue_sysfs_entry queue_nomerges_entry = {
219         .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
220         .show = queue_nomerges_show,
221         .store = queue_nomerges_store,
222 };
223
224 static struct queue_sysfs_entry queue_rq_affinity_entry = {
225         .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
226         .show = queue_rq_affinity_show,
227         .store = queue_rq_affinity_store,
228 };
229
230 static struct attribute *default_attrs[] = {
231         &queue_requests_entry.attr,
232         &queue_ra_entry.attr,
233         &queue_max_hw_sectors_entry.attr,
234         &queue_max_sectors_entry.attr,
235         &queue_iosched_entry.attr,
236         &queue_hw_sector_size_entry.attr,
237         &queue_nomerges_entry.attr,
238         &queue_rq_affinity_entry.attr,
239         NULL,
240 };
241
242 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
243
244 static ssize_t
245 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
246 {
247         struct queue_sysfs_entry *entry = to_queue(attr);
248         struct request_queue *q =
249                 container_of(kobj, struct request_queue, kobj);
250         ssize_t res;
251
252         if (!entry->show)
253                 return -EIO;
254         mutex_lock(&q->sysfs_lock);
255         if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
256                 mutex_unlock(&q->sysfs_lock);
257                 return -ENOENT;
258         }
259         res = entry->show(q, page);
260         mutex_unlock(&q->sysfs_lock);
261         return res;
262 }
263
264 static ssize_t
265 queue_attr_store(struct kobject *kobj, struct attribute *attr,
266                     const char *page, size_t length)
267 {
268         struct queue_sysfs_entry *entry = to_queue(attr);
269         struct request_queue *q;
270         ssize_t res;
271
272         if (!entry->store)
273                 return -EIO;
274
275         q = container_of(kobj, struct request_queue, kobj);
276         mutex_lock(&q->sysfs_lock);
277         if (test_bit(QUEUE_FLAG_DEAD, &q->queue_flags)) {
278                 mutex_unlock(&q->sysfs_lock);
279                 return -ENOENT;
280         }
281         res = entry->store(q, page, length);
282         mutex_unlock(&q->sysfs_lock);
283         return res;
284 }
285
286 /**
287  * blk_cleanup_queue: - release a &struct request_queue when it is no longer needed
288  * @kobj:    the kobj belonging of the request queue to be released
289  *
290  * Description:
291  *     blk_cleanup_queue is the pair to blk_init_queue() or
292  *     blk_queue_make_request().  It should be called when a request queue is
293  *     being released; typically when a block device is being de-registered.
294  *     Currently, its primary task it to free all the &struct request
295  *     structures that were allocated to the queue and the queue itself.
296  *
297  * Caveat:
298  *     Hopefully the low level driver will have finished any
299  *     outstanding requests first...
300  **/
301 static void blk_release_queue(struct kobject *kobj)
302 {
303         struct request_queue *q =
304                 container_of(kobj, struct request_queue, kobj);
305         struct request_list *rl = &q->rq;
306
307         blk_sync_queue(q);
308
309         if (rl->rq_pool)
310                 mempool_destroy(rl->rq_pool);
311
312         if (q->queue_tags)
313                 __blk_queue_free_tags(q);
314
315         blk_trace_shutdown(q);
316
317         bdi_destroy(&q->backing_dev_info);
318         kmem_cache_free(blk_requestq_cachep, q);
319 }
320
321 static struct sysfs_ops queue_sysfs_ops = {
322         .show   = queue_attr_show,
323         .store  = queue_attr_store,
324 };
325
326 struct kobj_type blk_queue_ktype = {
327         .sysfs_ops      = &queue_sysfs_ops,
328         .default_attrs  = default_attrs,
329         .release        = blk_release_queue,
330 };
331
332 int blk_register_queue(struct gendisk *disk)
333 {
334         int ret;
335
336         struct request_queue *q = disk->queue;
337
338         if (WARN_ON(!q))
339                 return -ENXIO;
340
341         if (!q->request_fn)
342                 return 0;
343
344         ret = kobject_add(&q->kobj, kobject_get(&disk_to_dev(disk)->kobj),
345                           "%s", "queue");
346         if (ret < 0)
347                 return ret;
348
349         kobject_uevent(&q->kobj, KOBJ_ADD);
350
351         ret = elv_register_queue(q);
352         if (ret) {
353                 kobject_uevent(&q->kobj, KOBJ_REMOVE);
354                 kobject_del(&q->kobj);
355                 return ret;
356         }
357
358         return 0;
359 }
360
361 void blk_unregister_queue(struct gendisk *disk)
362 {
363         struct request_queue *q = disk->queue;
364
365         if (WARN_ON(!q))
366                 return;
367
368         if (q->request_fn) {
369                 elv_unregister_queue(q);
370
371                 kobject_uevent(&q->kobj, KOBJ_REMOVE);
372                 kobject_del(&q->kobj);
373                 kobject_put(&disk_to_dev(disk)->kobj);
374         }
375 }