Re: Slow boot in QEMU with virtio-scsi disks

[Ming Lei]

On Sat, Aug 11, 2018 at 5:47 PM, Oleksandr Natalenko
<oleksandr@xxxxxxxxxxxxxx> wrote:
> Hi.
>
> I'd like to resurrect previous discussion [1] regarding slow kernel boot
> inside QEMU with virtio-scsi disks attached and blk_mq enabled.
>
> Symptom:
>
> [    2.830857] ata1: SATA max UDMA/133 abar m4096@0x98002000 port 0x98002100
> irq 36
> [    2.834559] ata2: SATA max UDMA/133 abar m4096@0x98002000 port 0x98002180
> irq 36
> [    2.837746] ata3: SATA max UDMA/133 abar m4096@0x98002000 port 0x98002200
> irq 36
> [    2.841861] ata4: SATA max UDMA/133 abar m4096@0x98002000 port 0x98002280
> irq 36
> [    2.847899] ata5: SATA max UDMA/133 abar m4096@0x98002000 port 0x98002300
> irq 36
> [    2.853229] ata6: SATA max UDMA/133 abar m4096@0x98002000 port 0x98002380
> irq 36
> [    3.172159] ata1: SATA link down (SStatus 0 SControl 300)
> [    3.183552] ata5: SATA link down (SStatus 0 SControl 300)
> [    3.189925] ata3: SATA link down (SStatus 0 SControl 300)
> [    3.196156] ata6: SATA link down (SStatus 0 SControl 300)
> [    3.201136] ata2: SATA link down (SStatus 0 SControl 300)
> [    3.208559] ata4: SATA link down (SStatus 0 SControl 300)
> [   16.480972] sd 0:0:1:0: Power-on or device reset occurred
> [   16.481591] sd 0:0:0:0: [sda] 16777216 512-byte logical blocks: (8.59
> GB/8.00 GiB)
> [   16.481671] sd 0:0:0:0: [sda] Write Protect is off
> [   16.481815] sd 0:0:0:0: [sda] Write cache: disabled, read cache: enabled,
> doesn't support DPO or FUA
> [   16.491325]  sda: sda1 sda2
> [   16.517532] sd 0:0:1:0: [sdb] 16777216 512-byte logical blocks: (8.59
> GB/8.00 GiB)
> [   16.525131] sr 0:0:2:0: Power-on or device reset occurred
> [   16.525974] sd 0:0:1:0: [sdb] Write Protect is off
> [   16.530946] sr 0:0:2:0: [sr0] scsi3-mmc drive: 16x/50x cd/rw xa/form2
> cdda tray
> [   16.543592] cdrom: Uniform CD-ROM driver Revision: 3.20
> [   16.549815] sd 0:0:1:0: [sdb] Write cache: disabled, read cache: enabled,
> doesn't support DPO or FUA
> [   16.549833] sd 0:0:0:0: [sda] Attached SCSI disk
> [   16.572055]  sdb: sdb1 sdb2
> [   16.580463] sd 0:0:1:0: [sdb] Attached SCSI disk
>
> (note the hang that lasts for 13 seconds)
>
> The disks are attached to the VM in the following manner:
>
> -device virtio-scsi,id=scsi -device scsi-hd,drive=hd1 -drive
> if=none,media=disk,id=hd1,file=sda.img,format=raw
>
> What I've tested so far:
>
> * 4.14.62  + virtio-scsi +    blk_mq == slow boot
> * 4.14.62  + virtio-scsi + no blk_mq == fast boot
> * 4.17.13  + virtio-scsi +    blk_mq == slow boot
> * 4.18-rc8 + virtio-scsi +    blk_mq == slow boot
>
> QEMU is of v2.12.1, runs with "-machine q35,accel=kvm -cpu host". Also, if
> virtio-scsi disks are replaced with SATA disks, the hang does not occur
> (although, QEMU has other issues with SATA, but that's another story [3]).
>
> Apparently, the commit that was mentioned in [2],
> b5b6e8c8d3b4cbeb447a0f10c7d5de3caa573299, forces blk_mq for virtio_scsi, so
> it cannot be disabled for new kernels.
>
> Any hint on how to avoid this hang while still having virtio-scsi disks and
> blk_mq enabled please?

Please test for-4.19/block:

https://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux-block.git/log/?h=for-4.19/block

This slow boot issue should have been fixed by the following commits:

1311326cf475 blk-mq: avoid to synchronize rcu inside blk_cleanup_queue()
97889f9ac24f blk-mq: remove synchronize_rcu() from blk_mq_del_queue_tag_set()
5815839b3ca1 blk-mq: introduce new lock for protecting hctx->dispatch_wait
2278d69f030f blk-mq: don't pass **hctx to blk_mq_mark_tag_wait()
8ab6bb9ee8d0 blk-mq: cleanup blk_mq_get_driver_tag()


Thanks,
Ming Lei