virtio_blk: allow blk-mq to merge scatter gather lists

As it's now possible to have arbitrary number of segments, mq-capable
block drivers should set BLK_MQ_F_SG_MERGE in q->queue_flags.
Especially virtio_blk must do it, __blk_segment_map_sg() may create
too large sg lists. When QUEUE_FLAG_NO_SG_MERGE turned on, such sg lists
could later trigger BUG_ON(total_sg > vq->vring.num) in virtqueue_add().

[   26.763457] ------------[ cut here ]------------
[   26.764147] kernel BUG at /home/mlin/linux/drivers/virtio/virtio_ring.c:160!
[   26.765010] invalid opcode: 0000 [#1] PREEMPT SMP
[   26.765273] Modules linked in:
[   26.765273] CPU: 3 PID: 25 Comm: kworker/u8:1 Not tainted 3.18.0-00030-g141405f torvalds#73
[   26.765273] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.7.5.1-0-g8936dbb-20141113_115728-nilsson.home.kraxel.org 04/01/2014
[   26.765273] Workqueue: writeback bdi_writeback_workfn (flush-254:0)
[   26.765273] task: ffff88001ebf0950 ti: ffff88001ebfc000 task.ti: ffff88001ebfc000
[   26.765273] RIP: 0010:[<ffffffff81314359>]  [<ffffffff81314359>] virtqueue_add_sgs+0x70/0x2fe
[   26.765273] RSP: 0000:ffff88001ebff6a8  EFLAGS: 00010002
[   26.765273] RAX: 0000000000000000 RBX: ffff88001e1ad000 RCX: ffffea0000698540
[   26.765273] RDX: ffffea0000698500 RSI: ffff88001ebff7c8 RDI: ffff88001ebff748
[   26.765273] RBP: ffff88001ebff708 R08: ffff88001d911330 R09: 0000000000000020
[   26.765273] R10: 0000160000000000 R11: ffff88001d7f0000 R12: ffff88001ebff7c8
[   26.765273] R13: 0000000000000002 R14: 0000000000000082 R15: 0000000000000003
[   26.777232] FS:  0000000000000000(0000) GS:ffff88001fd80000(0000) knlGS:0000000000000000
[   26.777232] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[   26.777232] CR2: 00007fd84f9e5000 CR3: 000000001cf6a000 CR4: 00000000000006e0
[   26.777232] Stack:
[   26.777232]  ffff88000000007f ffff880000000020 ffff88001ebff7d8 ffff88001d911330
[   26.777232]  01ffea0000000003 0000007f00000080 0000000000000418 0000000000000000
[   26.777232]  ffff88001ebff748 ffff88001d911360 0000000000000002 ffff88001d911330
[   26.777232] Call Trace:
[   26.777232]  [<ffffffff81346b3b>] __virtblk_add_req+0x137/0x149
[   26.777232]  [<ffffffff812c027e>] ? part_round_stats+0x52/0x59
[   26.777232]  [<ffffffff812c6c2f>] ? __blk_segment_map_sg+0xbd/0x190
[   26.777232]  [<ffffffff812c6e07>] ? blk_rq_map_sg+0xb6/0x1cc
[   26.777232]  [<ffffffff81346c9b>] virtio_queue_rq+0x14e/0x1ee
[   26.777232]  [<ffffffff812ca138>] __blk_mq_run_hw_queue+0x1ac/0x2ad
[   26.777232]  [<ffffffff812ca7f2>] blk_mq_run_hw_queue+0x3d/0x77
[   26.777232]  [<ffffffff812caf81>] blk_mq_insert_requests+0x103/0x156
[   26.777232]  [<ffffffff812cbdd5>] blk_mq_flush_plug_list+0xeb/0xfa
[   26.777232]  [<ffffffff812c365f>] blk_flush_plug_list+0xb8/0x1e1
[   26.777232]  [<ffffffff812c379e>] blk_finish_plug+0x16/0x38
[   26.777232]  [<ffffffff811af1f9>] ext4_writepages+0x973/0xb66
[   26.777232]  [<ffffffff810dd9ff>] do_writepages+0x1e/0x2c
[   26.777232]  [<ffffffff81145118>] __writeback_single_inode+0x84/0x298
[   26.777232]  [<ffffffff8106aa0d>] ? wake_up_bit+0x25/0x2a
[   26.777232]  [<ffffffff8114638f>] writeback_sb_inodes+0x1fc/0x33d
[   26.777232]  [<ffffffff81146544>] __writeback_inodes_wb+0x74/0xb9
[   26.777232]  [<ffffffff811466d5>] wb_writeback+0x14c/0x2eb
[   26.777232]  [<ffffffff810dc7b1>] ? global_dirty_limits+0x1b/0x117
[   26.777232]  [<ffffffff81146b92>] bdi_writeback_workfn+0x1f6/0x40c
[   26.777232]  [<ffffffff8104ded7>] process_one_work+0x1ca/0x376
[   26.777232]  [<ffffffff8104e319>] worker_thread+0x267/0x366
[   26.777232]  [<ffffffff8104e0b2>] ? process_scheduled_works+0x2f/0x2f
[   26.777232]  [<ffffffff81051e7c>] kthread+0xd2/0xda
[   26.777232]  [<ffffffff81490000>] ? ldsem_down_write+0x6/0x19f
[   26.777232]  [<ffffffff81051daa>] ? kthread_freezable_should_stop+0x48/0x48
[   26.777232]  [<ffffffff81490b6c>] ret_from_fork+0x7c/0xb0
[   26.777232]  [<ffffffff81051daa>] ? kthread_freezable_should_stop+0x48/0x48

Signed-off-by: Ming Lin <[email protected]>
[dpark: added a precise description]
Signed-off-by: Dongsu Park <[email protected]>

Author: Ming Lin

drivers/block/virtio_blk.c

@@ -626,7 +626,7 @@ static int virtblk_probe(struct virtio_device *vdev)
 	vblk->tag_set.ops = &virtio_mq_ops;
 	vblk->tag_set.queue_depth = virtblk_queue_depth;
 	vblk->tag_set.numa_node = NUMA_NO_NODE;
-	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
+	vblk->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 	vblk->tag_set.cmd_size =
 		sizeof(struct virtblk_req) +
 		sizeof(struct scatterlist) * sg_elems;