cfq-iosched: reduce bit operations in cfq_choose_req()
[linux-2.6.git] / block / blk-sysfs.c
index 26f9ec2..d935bd8 100644 (file)
@@ -2,6 +2,7 @@
  * Functions related to sysfs handling
  */
 #include <linux/kernel.h>
+#include <linux/slab.h>
 #include <linux/module.h>
 #include <linux/bio.h>
 #include <linux/blkdev.h>
@@ -16,9 +17,9 @@ struct queue_sysfs_entry {
 };
 
 static ssize_t
-queue_var_show(unsigned int var, char *page)
+queue_var_show(unsigned long var, char *page)
 {
-       return sprintf(page, "%d\n", var);
+       return sprintf(page, "%lu\n", var);
 }
 
 static ssize_t
@@ -40,7 +41,12 @@ queue_requests_store(struct request_queue *q, const char *page, size_t count)
 {
        struct request_list *rl = &q->rq;
        unsigned long nr;
-       int ret = queue_var_store(&nr, page, count);
+       int ret;
+
+       if (!q->request_fn)
+               return -EINVAL;
+
+       ret = queue_var_store(&nr, page, count);
        if (nr < BLKDEV_MIN_RQ)
                nr = BLKDEV_MIN_RQ;
 
@@ -60,14 +66,14 @@ queue_requests_store(struct request_queue *q, const char *page, size_t count)
 
        if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
                blk_set_queue_full(q, BLK_RW_SYNC);
-       } else if (rl->count[BLK_RW_SYNC]+1 <= q->nr_requests) {
+       } else {
                blk_clear_queue_full(q, BLK_RW_SYNC);
                wake_up(&rl->wait[BLK_RW_SYNC]);
        }
 
        if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
                blk_set_queue_full(q, BLK_RW_ASYNC);
-       } else if (rl->count[BLK_RW_ASYNC]+1 <= q->nr_requests) {
+       } else {
                blk_clear_queue_full(q, BLK_RW_ASYNC);
                wake_up(&rl->wait[BLK_RW_ASYNC]);
        }
@@ -77,7 +83,8 @@ queue_requests_store(struct request_queue *q, const char *page, size_t count)
 
 static ssize_t queue_ra_show(struct request_queue *q, char *page)
 {
-       int ra_kb = q->backing_dev_info.ra_pages << (PAGE_CACHE_SHIFT - 10);
+       unsigned long ra_kb = q->backing_dev_info.ra_pages <<
+                                       (PAGE_CACHE_SHIFT - 10);
 
        return queue_var_show(ra_kb, (page));
 }
@@ -95,21 +102,70 @@ queue_ra_store(struct request_queue *q, const char *page, size_t count)
 
 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
 {
-       int max_sectors_kb = q->max_sectors >> 1;
+       int max_sectors_kb = queue_max_sectors(q) >> 1;
 
        return queue_var_show(max_sectors_kb, (page));
 }
 
-static ssize_t queue_hw_sector_size_show(struct request_queue *q, char *page)
+static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(queue_max_segments(q), (page));
+}
+
+static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(q->limits.max_integrity_segments, (page));
+}
+
+static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 {
-       return queue_var_show(q->hardsect_size, page);
+       if (blk_queue_cluster(q))
+               return queue_var_show(queue_max_segment_size(q), (page));
+
+       return queue_var_show(PAGE_CACHE_SIZE, (page));
+}
+
+static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(queue_logical_block_size(q), page);
+}
+
+static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(queue_physical_block_size(q), page);
+}
+
+static ssize_t queue_io_min_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(queue_io_min(q), page);
+}
+
+static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(queue_io_opt(q), page);
+}
+
+static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(q->limits.discard_granularity, page);
+}
+
+static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
+{
+       return sprintf(page, "%llu\n",
+                      (unsigned long long)q->limits.max_discard_sectors << 9);
+}
+
+static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
+{
+       return queue_var_show(queue_discard_zeroes_data(q), page);
 }
 
 static ssize_t
 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 {
        unsigned long max_sectors_kb,
-                       max_hw_sectors_kb = q->max_hw_sectors >> 1,
+               max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
                        page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
        ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
 
@@ -117,7 +173,7 @@ queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
                return -EINVAL;
 
        spin_lock_irq(q->queue_lock);
-       q->max_sectors = max_sectors_kb << 1;
+       q->limits.max_sectors = max_sectors_kb << 1;
        spin_unlock_irq(q->queue_lock);
 
        return ret;
@@ -125,35 +181,46 @@ queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 
 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
 {
-       int max_hw_sectors_kb = q->max_hw_sectors >> 1;
+       int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
 
        return queue_var_show(max_hw_sectors_kb, (page));
 }
 
-static ssize_t queue_nonrot_show(struct request_queue *q, char *page)
-{
-       return queue_var_show(!blk_queue_nonrot(q), page);
+#define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                           \
+static ssize_t                                                         \
+queue_show_##name(struct request_queue *q, char *page)                 \
+{                                                                      \
+       int bit;                                                        \
+       bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
+       return queue_var_show(neg ? !bit : bit, page);                  \
+}                                                                      \
+static ssize_t                                                         \
+queue_store_##name(struct request_queue *q, const char *page, size_t count) \
+{                                                                      \
+       unsigned long val;                                              \
+       ssize_t ret;                                                    \
+       ret = queue_var_store(&val, page, count);                       \
+       if (neg)                                                        \
+               val = !val;                                             \
+                                                                       \
+       spin_lock_irq(q->queue_lock);                                   \
+       if (val)                                                        \
+               queue_flag_set(QUEUE_FLAG_##flag, q);                   \
+       else                                                            \
+               queue_flag_clear(QUEUE_FLAG_##flag, q);                 \
+       spin_unlock_irq(q->queue_lock);                                 \
+       return ret;                                                     \
 }
 
-static ssize_t queue_nonrot_store(struct request_queue *q, const char *page,
-                                 size_t count)
-{
-       unsigned long nm;
-       ssize_t ret = queue_var_store(&nm, page, count);
-
-       spin_lock_irq(q->queue_lock);
-       if (nm)
-               queue_flag_clear(QUEUE_FLAG_NONROT, q);
-       else
-               queue_flag_set(QUEUE_FLAG_NONROT, q);
-       spin_unlock_irq(q->queue_lock);
-
-       return ret;
-}
+QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
+QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
+QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
+#undef QUEUE_SYSFS_BIT_FNS
 
 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 {
-       return queue_var_show(blk_queue_nomerges(q), page);
+       return queue_var_show((blk_queue_nomerges(q) << 1) |
+                              blk_queue_noxmerges(q), page);
 }
 
 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
@@ -163,10 +230,12 @@ static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
        ssize_t ret = queue_var_store(&nm, page, count);
 
        spin_lock_irq(q->queue_lock);
-       if (nm)
+       queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
+       queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
+       if (nm == 2)
                queue_flag_set(QUEUE_FLAG_NOMERGES, q);
-       else
-               queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
+       else if (nm)
+               queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
        spin_unlock_irq(q->queue_lock);
 
        return ret;
@@ -174,9 +243,9 @@ static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 
 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 {
-       unsigned int set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
+       bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 
-       return queue_var_show(set != 0, page);
+       return queue_var_show(set, page);
 }
 
 static ssize_t
@@ -197,27 +266,6 @@ queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
        return ret;
 }
 
-static ssize_t queue_iostats_show(struct request_queue *q, char *page)
-{
-       return queue_var_show(blk_queue_io_stat(q), page);
-}
-
-static ssize_t queue_iostats_store(struct request_queue *q, const char *page,
-                                  size_t count)
-{
-       unsigned long stats;
-       ssize_t ret = queue_var_store(&stats, page, count);
-
-       spin_lock_irq(q->queue_lock);
-       if (stats)
-               queue_flag_set(QUEUE_FLAG_IO_STAT, q);
-       else
-               queue_flag_clear(QUEUE_FLAG_IO_STAT, q);
-       spin_unlock_irq(q->queue_lock);
-
-       return ret;
-}
-
 static struct queue_sysfs_entry queue_requests_entry = {
        .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
        .show = queue_requests_show,
@@ -241,6 +289,21 @@ static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
        .show = queue_max_hw_sectors_show,
 };
 
+static struct queue_sysfs_entry queue_max_segments_entry = {
+       .attr = {.name = "max_segments", .mode = S_IRUGO },
+       .show = queue_max_segments_show,
+};
+
+static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
+       .attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
+       .show = queue_max_integrity_segments_show,
+};
+
+static struct queue_sysfs_entry queue_max_segment_size_entry = {
+       .attr = {.name = "max_segment_size", .mode = S_IRUGO },
+       .show = queue_max_segment_size_show,
+};
+
 static struct queue_sysfs_entry queue_iosched_entry = {
        .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
        .show = elv_iosched_show,
@@ -249,13 +312,48 @@ static struct queue_sysfs_entry queue_iosched_entry = {
 
 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
        .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
-       .show = queue_hw_sector_size_show,
+       .show = queue_logical_block_size_show,
+};
+
+static struct queue_sysfs_entry queue_logical_block_size_entry = {
+       .attr = {.name = "logical_block_size", .mode = S_IRUGO },
+       .show = queue_logical_block_size_show,
+};
+
+static struct queue_sysfs_entry queue_physical_block_size_entry = {
+       .attr = {.name = "physical_block_size", .mode = S_IRUGO },
+       .show = queue_physical_block_size_show,
+};
+
+static struct queue_sysfs_entry queue_io_min_entry = {
+       .attr = {.name = "minimum_io_size", .mode = S_IRUGO },
+       .show = queue_io_min_show,
+};
+
+static struct queue_sysfs_entry queue_io_opt_entry = {
+       .attr = {.name = "optimal_io_size", .mode = S_IRUGO },
+       .show = queue_io_opt_show,
+};
+
+static struct queue_sysfs_entry queue_discard_granularity_entry = {
+       .attr = {.name = "discard_granularity", .mode = S_IRUGO },
+       .show = queue_discard_granularity_show,
+};
+
+static struct queue_sysfs_entry queue_discard_max_entry = {
+       .attr = {.name = "discard_max_bytes", .mode = S_IRUGO },
+       .show = queue_discard_max_show,
+};
+
+static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
+       .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
+       .show = queue_discard_zeroes_data_show,
 };
 
 static struct queue_sysfs_entry queue_nonrot_entry = {
        .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
-       .show = queue_nonrot_show,
-       .store = queue_nonrot_store,
+       .show = queue_show_nonrot,
+       .store = queue_store_nonrot,
 };
 
 static struct queue_sysfs_entry queue_nomerges_entry = {
@@ -272,8 +370,14 @@ static struct queue_sysfs_entry queue_rq_affinity_entry = {
 
 static struct queue_sysfs_entry queue_iostats_entry = {
        .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
-       .show = queue_iostats_show,
-       .store = queue_iostats_store,
+       .show = queue_show_iostats,
+       .store = queue_store_iostats,
+};
+
+static struct queue_sysfs_entry queue_random_entry = {
+       .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
+       .show = queue_show_random,
+       .store = queue_store_random,
 };
 
 static struct attribute *default_attrs[] = {
@@ -281,12 +385,23 @@ static struct attribute *default_attrs[] = {
        &queue_ra_entry.attr,
        &queue_max_hw_sectors_entry.attr,
        &queue_max_sectors_entry.attr,
+       &queue_max_segments_entry.attr,
+       &queue_max_integrity_segments_entry.attr,
+       &queue_max_segment_size_entry.attr,
        &queue_iosched_entry.attr,
        &queue_hw_sector_size_entry.attr,
+       &queue_logical_block_size_entry.attr,
+       &queue_physical_block_size_entry.attr,
+       &queue_io_min_entry.attr,
+       &queue_io_opt_entry.attr,
+       &queue_discard_granularity_entry.attr,
+       &queue_discard_max_entry.attr,
+       &queue_discard_zeroes_data_entry.attr,
        &queue_nonrot_entry.attr,
        &queue_nomerges_entry.attr,
        &queue_rq_affinity_entry.attr,
        &queue_iostats_entry.attr,
+       &queue_random_entry.attr,
        NULL,
 };
 
@@ -369,7 +484,7 @@ static void blk_release_queue(struct kobject *kobj)
        kmem_cache_free(blk_requestq_cachep, q);
 }
 
-static struct sysfs_ops queue_sysfs_ops = {
+static const struct sysfs_ops queue_sysfs_ops = {
        .show   = queue_attr_show,
        .store  = queue_attr_store,
 };
@@ -384,7 +499,6 @@ int blk_register_queue(struct gendisk *disk)
 {
        int ret;
        struct device *dev = disk_to_dev(disk);
-
        struct request_queue *q = disk->queue;
 
        if (WARN_ON(!q))
@@ -394,20 +508,23 @@ int blk_register_queue(struct gendisk *disk)
        if (ret)
                return ret;
 
-       if (!q->request_fn)
-               return 0;
-
-       ret = kobject_add(&q->kobj, kobject_get(&dev->kobj),
-                         "%s", "queue");
-       if (ret < 0)
+       ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
+       if (ret < 0) {
+               blk_trace_remove_sysfs(dev);
                return ret;
+       }
 
        kobject_uevent(&q->kobj, KOBJ_ADD);
 
+       if (!q->request_fn)
+               return 0;
+
        ret = elv_register_queue(q);
        if (ret) {
                kobject_uevent(&q->kobj, KOBJ_REMOVE);
                kobject_del(&q->kobj);
+               blk_trace_remove_sysfs(dev);
+               kobject_put(&dev->kobj);
                return ret;
        }
 
@@ -421,11 +538,11 @@ void blk_unregister_queue(struct gendisk *disk)
        if (WARN_ON(!q))
                return;
 
-       if (q->request_fn) {
+       if (q->request_fn)
                elv_unregister_queue(q);
 
-               kobject_uevent(&q->kobj, KOBJ_REMOVE);
-               kobject_del(&q->kobj);
-               kobject_put(&disk_to_dev(disk)->kobj);
-       }
+       kobject_uevent(&q->kobj, KOBJ_REMOVE);
+       kobject_del(&q->kobj);
+       blk_trace_remove_sysfs(disk_to_dev(disk));
+       kobject_put(&disk_to_dev(disk)->kobj);
 }