blk-mq: split blk_mq_alloc_and_init_hctx into two parts
authorMing Lei <ming.lei@redhat.com>
Tue, 30 Apr 2019 01:52:26 +0000 (09:52 +0800)
committerJens Axboe <axboe@kernel.dk>
Sat, 4 May 2019 13:24:06 +0000 (07:24 -0600)
Split blk_mq_alloc_and_init_hctx into two parts, and one is
blk_mq_alloc_hctx() for allocating all hctx resources, another
is blk_mq_init_hctx() for initializing hctx, which serves as
counter-part of blk_mq_exit_hctx().

Cc: Dongli Zhang <dongli.zhang@oracle.com>
Cc: James Smart <james.smart@broadcom.com>
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: linux-scsi@vger.kernel.org
Cc: Martin K . Petersen <martin.petersen@oracle.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: James E . J . Bottomley <jejb@linux.vnet.ibm.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-mq.c

index 1fdb8de92a10d2ef1f5c14eacdfcd13e03120ede..17e63d80b6d6020f41b661469a0aa98bb85044cc 100644 (file)
@@ -2285,15 +2285,65 @@ static void blk_mq_exit_hw_queues(struct request_queue *q,
        }
 }
 
+static int blk_mq_hw_ctx_size(struct blk_mq_tag_set *tag_set)
+{
+       int hw_ctx_size = sizeof(struct blk_mq_hw_ctx);
+
+       BUILD_BUG_ON(ALIGN(offsetof(struct blk_mq_hw_ctx, srcu),
+                          __alignof__(struct blk_mq_hw_ctx)) !=
+                    sizeof(struct blk_mq_hw_ctx));
+
+       if (tag_set->flags & BLK_MQ_F_BLOCKING)
+               hw_ctx_size += sizeof(struct srcu_struct);
+
+       return hw_ctx_size;
+}
+
 static int blk_mq_init_hctx(struct request_queue *q,
                struct blk_mq_tag_set *set,
                struct blk_mq_hw_ctx *hctx, unsigned hctx_idx)
 {
-       int node;
+       hctx->queue_num = hctx_idx;
+
+       cpuhp_state_add_instance_nocalls(CPUHP_BLK_MQ_DEAD, &hctx->cpuhp_dead);
+
+       hctx->tags = set->tags[hctx_idx];
+
+       if (set->ops->init_hctx &&
+           set->ops->init_hctx(hctx, set->driver_data, hctx_idx))
+               goto unregister_cpu_notifier;
 
-       node = hctx->numa_node;
+       if (blk_mq_init_request(set, hctx->fq->flush_rq, hctx_idx,
+                               hctx->numa_node))
+               goto exit_hctx;
+       return 0;
+
+ exit_hctx:
+       if (set->ops->exit_hctx)
+               set->ops->exit_hctx(hctx, hctx_idx);
+ unregister_cpu_notifier:
+       blk_mq_remove_cpuhp(hctx);
+       return -1;
+}
+
+static struct blk_mq_hw_ctx *
+blk_mq_alloc_hctx(struct request_queue *q, struct blk_mq_tag_set *set,
+               int node)
+{
+       struct blk_mq_hw_ctx *hctx;
+       gfp_t gfp = GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY;
+
+       hctx = kzalloc_node(blk_mq_hw_ctx_size(set), gfp, node);
+       if (!hctx)
+               goto fail_alloc_hctx;
+
+       if (!zalloc_cpumask_var_node(&hctx->cpumask, gfp, node))
+               goto free_hctx;
+
+       atomic_set(&hctx->nr_active, 0);
        if (node == NUMA_NO_NODE)
-               node = hctx->numa_node = set->numa_node;
+               node = set->numa_node;
+       hctx->numa_node = node;
 
        INIT_DELAYED_WORK(&hctx->run_work, blk_mq_run_work_fn);
        spin_lock_init(&hctx->lock);
@@ -2301,58 +2351,45 @@ static int blk_mq_init_hctx(struct request_queue *q,
        hctx->queue = q;
        hctx->flags = set->flags & ~BLK_MQ_F_TAG_SHARED;
 
-       cpuhp_state_add_instance_nocalls(CPUHP_BLK_MQ_DEAD, &hctx->cpuhp_dead);
-
-       hctx->tags = set->tags[hctx_idx];
-
        /*
         * Allocate space for all possible cpus to avoid allocation at
         * runtime
         */
        hctx->ctxs = kmalloc_array_node(nr_cpu_ids, sizeof(void *),
-                       GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY, node);
+                       gfp, node);
        if (!hctx->ctxs)
-               goto unregister_cpu_notifier;
+               goto free_cpumask;
 
        if (sbitmap_init_node(&hctx->ctx_map, nr_cpu_ids, ilog2(8),
-                               GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY, node))
+                               gfp, node))
                goto free_ctxs;
-
        hctx->nr_ctx = 0;
 
        spin_lock_init(&hctx->dispatch_wait_lock);
        init_waitqueue_func_entry(&hctx->dispatch_wait, blk_mq_dispatch_wake);
        INIT_LIST_HEAD(&hctx->dispatch_wait.entry);
 
-       if (set->ops->init_hctx &&
-           set->ops->init_hctx(hctx, set->driver_data, hctx_idx))
-               goto free_bitmap;
-
        hctx->fq = blk_alloc_flush_queue(q, hctx->numa_node, set->cmd_size,
-                       GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY);
+                       gfp);
        if (!hctx->fq)
-               goto exit_hctx;
-
-       if (blk_mq_init_request(set, hctx->fq->flush_rq, hctx_idx, node))
-               goto free_fq;
+               goto free_bitmap;
 
        if (hctx->flags & BLK_MQ_F_BLOCKING)
                init_srcu_struct(hctx->srcu);
+       blk_mq_hctx_kobj_init(hctx);
 
-       return 0;
+       return hctx;
 
- free_fq:
-       blk_free_flush_queue(hctx->fq);
- exit_hctx:
-       if (set->ops->exit_hctx)
-               set->ops->exit_hctx(hctx, hctx_idx);
  free_bitmap:
        sbitmap_free(&hctx->ctx_map);
  free_ctxs:
        kfree(hctx->ctxs);
- unregister_cpu_notifier:
-       blk_mq_remove_cpuhp(hctx);
-       return -1;
+ free_cpumask:
+       free_cpumask_var(hctx->cpumask);
+ free_hctx:
+       kfree(hctx);
+ fail_alloc_hctx:
+       return NULL;
 }
 
 static void blk_mq_init_cpu_queues(struct request_queue *q,
@@ -2698,51 +2735,25 @@ struct request_queue *blk_mq_init_sq_queue(struct blk_mq_tag_set *set,
 }
 EXPORT_SYMBOL(blk_mq_init_sq_queue);
 
-static int blk_mq_hw_ctx_size(struct blk_mq_tag_set *tag_set)
-{
-       int hw_ctx_size = sizeof(struct blk_mq_hw_ctx);
-
-       BUILD_BUG_ON(ALIGN(offsetof(struct blk_mq_hw_ctx, srcu),
-                          __alignof__(struct blk_mq_hw_ctx)) !=
-                    sizeof(struct blk_mq_hw_ctx));
-
-       if (tag_set->flags & BLK_MQ_F_BLOCKING)
-               hw_ctx_size += sizeof(struct srcu_struct);
-
-       return hw_ctx_size;
-}
-
 static struct blk_mq_hw_ctx *blk_mq_alloc_and_init_hctx(
                struct blk_mq_tag_set *set, struct request_queue *q,
                int hctx_idx, int node)
 {
        struct blk_mq_hw_ctx *hctx;
 
-       hctx = kzalloc_node(blk_mq_hw_ctx_size(set),
-                       GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY,
-                       node);
+       hctx = blk_mq_alloc_hctx(q, set, node);
        if (!hctx)
-               return NULL;
-
-       if (!zalloc_cpumask_var_node(&hctx->cpumask,
-                               GFP_NOIO | __GFP_NOWARN | __GFP_NORETRY,
-                               node)) {
-               kfree(hctx);
-               return NULL;
-       }
-
-       atomic_set(&hctx->nr_active, 0);
-       hctx->numa_node = node;
-       hctx->queue_num = hctx_idx;
+               goto fail;
 
-       if (blk_mq_init_hctx(q, set, hctx, hctx_idx)) {
-               free_cpumask_var(hctx->cpumask);
-               kfree(hctx);
-               return NULL;
-       }
-       blk_mq_hctx_kobj_init(hctx);
+       if (blk_mq_init_hctx(q, set, hctx, hctx_idx))
+               goto free_hctx;
 
        return hctx;
+
+ free_hctx:
+       kobject_put(&hctx->kobj);
+ fail:
+       return NULL;
 }
 
 static void blk_mq_realloc_hw_ctxs(struct blk_mq_tag_set *set,