Re: [HANG 3.13-rc3] blk-mq/virtio: mkfs.ext4 hangs in blk_mq_wait_for_tags

[Jens Axboe <axboe@kernel.dk>]

On 12/09/2013 03:39 PM, Dave Chinner wrote:
> Hi Jens,
> 
> Another day, another blkmq/virtio problem. Running mkfs.ext4 on a
> sparse 100TB VM file image, it hangs hard while writing superblock
> information:
> 
> $ tests/fsmark-50-test-ext4.sh
> mke2fs 1.43-WIP (20-Jun-2013)
> Filesystem label=
> OS type: Linux
> Block size=4096 (log=2)
> Fragment size=4096 (log=2)
> Stride=0 blocks, Stripe width=0 blocks
> 1677721600 inodes, 26843545600 blocks
> 1342177280 blocks (5.00%) reserved for the super user
> First data block=0
> 819200 block groups
> 32768 blocks per group, 32768 fragments per group
> 2048 inodes per group
> Superblock backups stored on blocks:
>         32768, 98304, 163840, 229376, 294912, 819200, 884736, 1605632, 2654208,
>         4096000, 7962624, 11239424, 20480000, 23887872, 71663616, 78675968,
>         102400000, 214990848, 512000000, 550731776, 644972544, 1934917632,
>         2560000000, 3855122432, 5804752896, 12800000000, 17414258688
> 
> Allocating group tables: done
> Writing inode tables: done
> Creating journal (32768 blocks): done
> Writing superblocks and filesystem accounting information: 
> 
> It writes a few superblocks, then hangs. Immediately after it stops
> updating that last line, I see this:
> 
> root@test4:~# echo w > /proc/sysrq-trigger 
> [   79.408153] SysRq : Show Blocked State
> [   79.408832]   task                        PC stack   pid father
> [   79.409860] mke2fs          D ffff88011bc13100  3904  4242   4241 0x00000002
> [   79.411074]  ffff88021a737978 0000000000000086 ffff8800dbb9de40 0000000000013100
> [   79.412009]  ffff88021a737fd8 0000000000013100 ffff88011ac7af20 ffff8800dbb9de40
> [   79.412009]  ffff88021a737988 ffffe8fcfbc038d0 ffff88011b39c058 ffff88011b39c040
> [   79.412009] Call Trace:
> [   79.412009]  [<ffffffff81ae36d9>] schedule+0x29/0x70
> [   79.412009]  [<ffffffff8178863e>] percpu_ida_alloc+0x16e/0x330
> [   79.412009]  [<ffffffff810cf393>] ? finish_wait+0x63/0x80
> [   79.412009]  [<ffffffff810cf3f0>] ? __init_waitqueue_head+0x40/0x40
> [   79.412009]  [<ffffffff8175f30f>] blk_mq_wait_for_tags+0x1f/0x40
> [   79.412009]  [<ffffffff8175e28e>] blk_mq_alloc_request_pinned+0x4e/0x110
> [   79.412009]  [<ffffffff8175eacb>] blk_mq_make_request+0x41b/0x500
> [   79.412009]  [<ffffffff81753552>] generic_make_request+0xc2/0x110
> [   79.412009]  [<ffffffff81754a1c>] submit_bio+0x6c/0x120
> [   79.412009]  [<ffffffff811d1dd3>] _submit_bh+0x133/0x200
> [   79.412009]  [<ffffffff811d1eb0>] submit_bh+0x10/0x20
> [   79.412009]  [<ffffffff811d5298>] __block_write_full_page+0x1b8/0x370
> [   79.412009]  [<ffffffff811d3e30>] ? block_read_full_page+0x320/0x320
> [   79.412009]  [<ffffffff811d8450>] ? I_BDEV+0x10/0x10
> [   79.412009]  [<ffffffff811d8450>] ? I_BDEV+0x10/0x10
> [   79.412009]  [<ffffffff811d5541>] block_write_full_page_endio+0xf1/0x100
> [   79.412009]  [<ffffffff811d5565>] block_write_full_page+0x15/0x20
> [   79.412009]  [<ffffffff811d8908>] blkdev_writepage+0x18/0x20
> [   79.412009]  [<ffffffff8115668a>] __writepage+0x1a/0x50
> [   79.412009]  [<ffffffff81157055>] write_cache_pages+0x225/0x470
> [   79.412009]  [<ffffffff81156670>] ? mapping_tagged+0x20/0x20
> [   79.412009]  [<ffffffff811572ed>] generic_writepages+0x4d/0x70
> [   79.412009]  [<ffffffff810c4d0f>] ? __dequeue_entity+0x2f/0x50
> [   79.412009]  [<ffffffff81158bd1>] do_writepages+0x21/0x50
> [   79.412009]  [<ffffffff8114e199>] __filemap_fdatawrite_range+0x59/0x60
> [   79.412009]  [<ffffffff81ae7e8e>] ? _raw_spin_unlock_irq+0xe/0x20
> [   79.412009]  [<ffffffff8114e1da>] filemap_write_and_wait_range+0x3a/0x80
> [   79.412009]  [<ffffffff811d8b14>] blkdev_fsync+0x24/0x50
> [   79.412009]  [<ffffffff811cf898>] do_fsync+0x58/0x80
> [   79.412009]  [<ffffffff81aeb8e5>] ? do_async_page_fault+0x35/0xc0
> [   79.412009]  [<ffffffff811cfb30>] SyS_fsync+0x10/0x20
> [   79.412009]  [<ffffffff81af08e9>] system_call_fastpath+0x16/0x1b
> 
> And a coupel of seconds later the VM hangs hard - console,
> networking, everything just stops dead and it doesn't even respond
> ot an nmi from the qemu command console.
> 
> The test is exactly the same as described in the previous problem I
> had:
> 
> http://marc.info/?l=linux-kernel&m=138621901319333&w=2
> 
> The only difference is that I'm trying to run the concurrent create
> workload on ext4 now, not XFS, and it's failing in mkfs.ext4 during
> the setup code....

I'll take a look at this.

> At this point, I have to ask: is anyone doing high IOPS testing on
> virtio/blk_mq? This is the third regression I've hit since it was
> merged, and I'm really not stressing this code nearly as much as
> some of the hardware out there is capable of doing....

Plenty was done previously, but at this point it does seem flakey.
Wonder if it's other changes, or some screwup along the way.

-- 
Jens Axboe