blk-mq-tag: cleanup the normal/reserved tag allocation

Hello Jens Axboe,

This is a semi-automatic email about new static checker warnings.

The patch 4941115bef2b: "blk-mq-tag: cleanup the normal/reserved tag allocation" from Jan 13, 2017, leads to the following Smatch complaint:

block/blk-mq-tag.c:142 blk_mq_get_tag()

	 warn: variable dereferenced before check 'data->hctx' (see line 102)

block/blk-mq-tag.c
101 {

   102		struct blk_mq_tags *tags = blk_mq_tags_from_data(data);
                                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^

We added a new "data->hctx" dereference here.

   103		struct sbitmap_queue *bt;
   104		struct sbq_wait_state *ws;
   105		DEFINE_WAIT(wait);
   106		unsigned int tag_offset;
   107		int tag;
   108	
   109		if (data->flags & BLK_MQ_REQ_RESERVED) {
   110			if (unlikely(!tags->nr_reserved_tags)) {
   111				WARN_ON_ONCE(1);
   112				return BLK_MQ_TAG_FAIL;
   113			}
   114			bt = &tags->breserved_tags;
   115			tag_offset = 0;
   116		} else {
   117			bt = &tags->bitmap_tags;
   118			tag_offset = tags->nr_reserved_tags;
   119		}
   120	
   121		tag = __blk_mq_get_tag(data->hctx, bt);
   122		if (tag != -1)
   123			goto found_tag;
   124	
   125		if (data->flags & BLK_MQ_REQ_NOWAIT)
   126			return BLK_MQ_TAG_FAIL;
   127	
   128		ws = bt_wait_ptr(bt, data->hctx);
   129		do {
   130			prepare_to_wait(&ws->wait, &wait, TASK_UNINTERRUPTIBLE);
   131	
   132			tag = __blk_mq_get_tag(data->hctx, bt);
   133			if (tag != -1)
   134				break;
   135	
   136			/*
   137			 * We're out of tags on this hardware queue, kick any
   138			 * pending IO submits before going to sleep waiting for
   139			 * some to complete. Note that hctx can be NULL here for
   140			 * reserved tag allocation.
   141			 */
   142			if (data->hctx)

We're in a loop here so it's possible we don't need to check the first item in the list, but it's kind of new so I thought I would forward the message anyway even though it might be a false positive.

   143				blk_mq_run_hw_queue(data->hctx, false);
   144	

regards,
dan carpenter

It's no longer possible for 'hctx' to be passed in as NULL, so the above dereference is safe. I'll tend to this superfluous hctx check.

Jens Axboe