Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 1 | #include <linux/kernel.h> |
| 2 | #include <linux/module.h> |
| 3 | #include <linux/percpu_ida.h> |
| 4 | |
| 5 | #include <linux/blk-mq.h> |
| 6 | #include "blk.h" |
| 7 | #include "blk-mq.h" |
| 8 | #include "blk-mq-tag.h" |
| 9 | |
| 10 | /* |
| 11 | * Per tagged queue (tag address space) map |
| 12 | */ |
| 13 | struct blk_mq_tags { |
| 14 | unsigned int nr_tags; |
| 15 | unsigned int nr_reserved_tags; |
| 16 | unsigned int nr_batch_move; |
| 17 | unsigned int nr_max_cache; |
| 18 | |
| 19 | struct percpu_ida free_tags; |
| 20 | struct percpu_ida reserved_tags; |
| 21 | }; |
| 22 | |
| 23 | void blk_mq_wait_for_tags(struct blk_mq_tags *tags) |
| 24 | { |
| 25 | int tag = blk_mq_get_tag(tags, __GFP_WAIT, false); |
| 26 | blk_mq_put_tag(tags, tag); |
| 27 | } |
| 28 | |
| 29 | bool blk_mq_has_free_tags(struct blk_mq_tags *tags) |
| 30 | { |
| 31 | return !tags || |
| 32 | percpu_ida_free_tags(&tags->free_tags, nr_cpu_ids) != 0; |
| 33 | } |
| 34 | |
| 35 | static unsigned int __blk_mq_get_tag(struct blk_mq_tags *tags, gfp_t gfp) |
| 36 | { |
| 37 | int tag; |
| 38 | |
Kent Overstreet | 6f6b5d1 | 2014-01-19 08:26:37 +0000 | [diff] [blame] | 39 | tag = percpu_ida_alloc(&tags->free_tags, (gfp & __GFP_WAIT) ? |
| 40 | TASK_UNINTERRUPTIBLE : TASK_RUNNING); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 41 | if (tag < 0) |
| 42 | return BLK_MQ_TAG_FAIL; |
| 43 | return tag + tags->nr_reserved_tags; |
| 44 | } |
| 45 | |
| 46 | static unsigned int __blk_mq_get_reserved_tag(struct blk_mq_tags *tags, |
| 47 | gfp_t gfp) |
| 48 | { |
| 49 | int tag; |
| 50 | |
| 51 | if (unlikely(!tags->nr_reserved_tags)) { |
| 52 | WARN_ON_ONCE(1); |
| 53 | return BLK_MQ_TAG_FAIL; |
| 54 | } |
| 55 | |
Kent Overstreet | 6f6b5d1 | 2014-01-19 08:26:37 +0000 | [diff] [blame] | 56 | tag = percpu_ida_alloc(&tags->reserved_tags, (gfp & __GFP_WAIT) ? |
| 57 | TASK_UNINTERRUPTIBLE : TASK_RUNNING); |
Jens Axboe | 320ae51 | 2013-10-24 09:20:05 +0100 | [diff] [blame] | 58 | if (tag < 0) |
| 59 | return BLK_MQ_TAG_FAIL; |
| 60 | return tag; |
| 61 | } |
| 62 | |
| 63 | unsigned int blk_mq_get_tag(struct blk_mq_tags *tags, gfp_t gfp, bool reserved) |
| 64 | { |
| 65 | if (!reserved) |
| 66 | return __blk_mq_get_tag(tags, gfp); |
| 67 | |
| 68 | return __blk_mq_get_reserved_tag(tags, gfp); |
| 69 | } |
| 70 | |
| 71 | static void __blk_mq_put_tag(struct blk_mq_tags *tags, unsigned int tag) |
| 72 | { |
| 73 | BUG_ON(tag >= tags->nr_tags); |
| 74 | |
| 75 | percpu_ida_free(&tags->free_tags, tag - tags->nr_reserved_tags); |
| 76 | } |
| 77 | |
| 78 | static void __blk_mq_put_reserved_tag(struct blk_mq_tags *tags, |
| 79 | unsigned int tag) |
| 80 | { |
| 81 | BUG_ON(tag >= tags->nr_reserved_tags); |
| 82 | |
| 83 | percpu_ida_free(&tags->reserved_tags, tag); |
| 84 | } |
| 85 | |
| 86 | void blk_mq_put_tag(struct blk_mq_tags *tags, unsigned int tag) |
| 87 | { |
| 88 | if (tag >= tags->nr_reserved_tags) |
| 89 | __blk_mq_put_tag(tags, tag); |
| 90 | else |
| 91 | __blk_mq_put_reserved_tag(tags, tag); |
| 92 | } |
| 93 | |
| 94 | static int __blk_mq_tag_iter(unsigned id, void *data) |
| 95 | { |
| 96 | unsigned long *tag_map = data; |
| 97 | __set_bit(id, tag_map); |
| 98 | return 0; |
| 99 | } |
| 100 | |
| 101 | void blk_mq_tag_busy_iter(struct blk_mq_tags *tags, |
| 102 | void (*fn)(void *, unsigned long *), void *data) |
| 103 | { |
| 104 | unsigned long *tag_map; |
| 105 | size_t map_size; |
| 106 | |
| 107 | map_size = ALIGN(tags->nr_tags, BITS_PER_LONG) / BITS_PER_LONG; |
| 108 | tag_map = kzalloc(map_size * sizeof(unsigned long), GFP_ATOMIC); |
| 109 | if (!tag_map) |
| 110 | return; |
| 111 | |
| 112 | percpu_ida_for_each_free(&tags->free_tags, __blk_mq_tag_iter, tag_map); |
| 113 | if (tags->nr_reserved_tags) |
| 114 | percpu_ida_for_each_free(&tags->reserved_tags, __blk_mq_tag_iter, |
| 115 | tag_map); |
| 116 | |
| 117 | fn(data, tag_map); |
| 118 | kfree(tag_map); |
| 119 | } |
| 120 | |
| 121 | struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags, |
| 122 | unsigned int reserved_tags, int node) |
| 123 | { |
| 124 | unsigned int nr_tags, nr_cache; |
| 125 | struct blk_mq_tags *tags; |
| 126 | int ret; |
| 127 | |
| 128 | if (total_tags > BLK_MQ_TAG_MAX) { |
| 129 | pr_err("blk-mq: tag depth too large\n"); |
| 130 | return NULL; |
| 131 | } |
| 132 | |
| 133 | tags = kzalloc_node(sizeof(*tags), GFP_KERNEL, node); |
| 134 | if (!tags) |
| 135 | return NULL; |
| 136 | |
| 137 | nr_tags = total_tags - reserved_tags; |
| 138 | nr_cache = nr_tags / num_possible_cpus(); |
| 139 | |
| 140 | if (nr_cache < BLK_MQ_TAG_CACHE_MIN) |
| 141 | nr_cache = BLK_MQ_TAG_CACHE_MIN; |
| 142 | else if (nr_cache > BLK_MQ_TAG_CACHE_MAX) |
| 143 | nr_cache = BLK_MQ_TAG_CACHE_MAX; |
| 144 | |
| 145 | tags->nr_tags = total_tags; |
| 146 | tags->nr_reserved_tags = reserved_tags; |
| 147 | tags->nr_max_cache = nr_cache; |
| 148 | tags->nr_batch_move = max(1u, nr_cache / 2); |
| 149 | |
| 150 | ret = __percpu_ida_init(&tags->free_tags, tags->nr_tags - |
| 151 | tags->nr_reserved_tags, |
| 152 | tags->nr_max_cache, |
| 153 | tags->nr_batch_move); |
| 154 | if (ret) |
| 155 | goto err_free_tags; |
| 156 | |
| 157 | if (reserved_tags) { |
| 158 | /* |
| 159 | * With max_cahe and batch set to 1, the allocator fallbacks to |
| 160 | * no cached. It's fine reserved tags allocation is slow. |
| 161 | */ |
| 162 | ret = __percpu_ida_init(&tags->reserved_tags, reserved_tags, |
| 163 | 1, 1); |
| 164 | if (ret) |
| 165 | goto err_reserved_tags; |
| 166 | } |
| 167 | |
| 168 | return tags; |
| 169 | |
| 170 | err_reserved_tags: |
| 171 | percpu_ida_destroy(&tags->free_tags); |
| 172 | err_free_tags: |
| 173 | kfree(tags); |
| 174 | return NULL; |
| 175 | } |
| 176 | |
| 177 | void blk_mq_free_tags(struct blk_mq_tags *tags) |
| 178 | { |
| 179 | percpu_ida_destroy(&tags->free_tags); |
| 180 | percpu_ida_destroy(&tags->reserved_tags); |
| 181 | kfree(tags); |
| 182 | } |
| 183 | |
| 184 | ssize_t blk_mq_tag_sysfs_show(struct blk_mq_tags *tags, char *page) |
| 185 | { |
| 186 | char *orig_page = page; |
| 187 | int cpu; |
| 188 | |
| 189 | if (!tags) |
| 190 | return 0; |
| 191 | |
| 192 | page += sprintf(page, "nr_tags=%u, reserved_tags=%u, batch_move=%u," |
| 193 | " max_cache=%u\n", tags->nr_tags, tags->nr_reserved_tags, |
| 194 | tags->nr_batch_move, tags->nr_max_cache); |
| 195 | |
| 196 | page += sprintf(page, "nr_free=%u, nr_reserved=%u\n", |
| 197 | percpu_ida_free_tags(&tags->free_tags, nr_cpu_ids), |
| 198 | percpu_ida_free_tags(&tags->reserved_tags, nr_cpu_ids)); |
| 199 | |
| 200 | for_each_possible_cpu(cpu) { |
| 201 | page += sprintf(page, " cpu%02u: nr_free=%u\n", cpu, |
| 202 | percpu_ida_free_tags(&tags->free_tags, cpu)); |
| 203 | } |
| 204 | |
| 205 | return page - orig_page; |
| 206 | } |