Re: Regression: commit 045065d breaks kernel on machine with atapi floppy: high IOWAIT, hung processes (bisected)

[Sergio Callegari <sergio.callegari@unibo.it>]

Seems that the issue also affects other systems with different configs:

https://bbs.archlinux.org/viewtopic.php?id=189324

Possibly, the same bug reported in

https://bugzilla.kernel.org/show_bug.cgi?id=87581

A tentative patch was submitted on LKML

https://lkml.org/lkml/2014/11/20/581

I have not tested it yet.

Another possible solution being reported is increasing delay time in 
blk_delay_queue(q, SCSI_QUEUE_DELAY)

Not tested yet either.

Threads in 189324 suggests that bug is triggered by mixing a slower 
device with a faster one on the same IDE/SATA channel.

Can someone indicate:

- If one of the two patches has already been accepted in recent kernels 
or is pending acceptance?

- Which one among the two approaches (extending delay time or modifying 
spin locks in scsi_lib.c) is more appropriate for me to test?

Best,

Sergio



On 16/08/2015 16:19, Sergio Callegari wrote:
>
> Hi,
>
> please keep me in CC in answers.
>
> I'd like to report that after commit
>
> [045065d8a300a37218c548e9aa7becd581c6a0e8] [SCSI] fix qemu boot hang
> problem
>
> the kernel is not usable on a machine with an IOMEGA Zip 100 ATAPI drive
> as in:
>
> Model=IOMEGA ZIP 100 ATAPI Floppy, FwRev=12.A, SerialNo=
> Config={ SpinMotCtl Removeable nonMagnetic }
> RawCHS=0/0/0, TrkSize=0, SectSize=0, ECCbytes=0
> BuffType=unknown, BuffSize=unknown, MaxMultSect=0 (maybe): CurCHS=0/0/0,
> CurSects=0, LBA=yes, LBAsects=0
> IORDY=on/off, tPIO={min:500,w/IORDY:180}
> PIO modes: pio0 pio1 pio2 pio3
> AdvancedPM=no
>
> Symptoms include:
>
> - Extremely high IOWAIT in absence of load
> - Kernel reporting hung processes
> - Commands like blkid hanging
> - Inability of the machine to shutdown
>
> Symptoms do not appear immediately, but after some time (anywhere
> between a few minutes and /many hours/ after boot). First symptom is
> IOWAIT suddendly jumping high.
>
> Due to the delay in which symptoms manifest, bisecting has been quite
> painful, but I am now rather sure that the first bad commit is the one
> above.
>
> Other pieces of hardware configuration include:
>
> - ASRock N68S motherboard with AMD Phenom(tm) II X4 920 Processor and
> NVIDIA MCP61 SATA/IDE Chipset
> - IDE drive connected as slave on ide interface where master is HL-DT-ST
> DVD-RAM GH22NP20 CDROM/DVD writer
>
> Issue is weird because the commit seems to merely fix a trivial error in
> logic condition
>
> -       if (atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
> +       if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
>                  blk_delay_queue(q, SCSI_QUEUE_DELAY);
>
> Hence, the commit may just end up making visible some other issue.
>
> Best,
>
> Sergio
>
>