cdparanoia 10.2 crashes system during drive detect

[Glenn Golden]

Fellow 'noids,

I recently attempted to upgrade from cdparanoia 9.8 (which I've been using
successfully for years on a variety of systems) to 10.2.  Alas, 10.2 seems
to be causing hard, 100% reproducible crashes during drive detection, even
running the static binary from the xiph website.

I've done my homework on the website (bug reports, "troubleshooting" section,
mailing list) and googled of course, but have not seen any mention of this.

To display the behavior under "fair" (i.e., identically-built) conditions,
I downloaded the static binaries for both 9.8 and 10.2, and ran them both.
Results comparison shown below, along with some random commentary at the end.

I will be happy to send more detailed system info or perform any additional
experiments that you wish in order to help diagnose the problem.

Thanks for any help, and thanks for creating cdparanoia in the first place.
It's a superb tool, and IMHO exemplifies what great free SW is all about.

Regards,

Glenn Golden


---------------
Behavior of cdparanoia-III-alpha9.8.i386-linux-elf -vsQ (static version
from xiph website), executed as normal user:

    nu$> ./cdparanoia-III-alpha9.8.i386-linux-elf -vsQ
    cdparanoia III release 9.8 (March 23, 2001)
    (C) 2001 Monty <monty@...> and Xiphophorus

    Report bugs to paranoia@...
    http://www.xiph.org/paranoia/

    Checking /dev/cdrom for cdrom...
            Testing /dev/cdrom for cooked ioctl() interface
                    /dev/scd0 is not a cooked ioctl CDROM.
            Testing /dev/cdrom for SCSI interface
                    generic device: /dev/sg1
                    ioctl device: /dev/scd0

    Found an accessible SCSI CDROM drive.
    Looking at revision of the SG interface in use...
            SG interface version 3.5.33; OK.

    CDROM model sensed sensed: HL-DT-ST RW/DVD GCC-4242N 0J05

    Checking for SCSI emulation...
            Drive is ATAPI (using SCSI host adaptor emulation)
            Couldn't disable kernel command translation layer

    Checking for MMC style command set...
            Drive is MMC style
            DMA scatter/gather table entries: 127
            table entry size: 32768 bytes
            maximum theoretical transfer: 1769 sectors
            Setting default read size to 13 sectors (30576 bytes).

    Verifying CDDA command set...
            Expected command set reads OK.

    Table of contents (audio tracks only):
        .
        .
        .

Operation is normal from this point.  The only unusual behavior are numerous
warnings like this in the syslog while doing rips:

    sg_write: data in/out 26/26 bytes for SCSI command 0x5a--guessing data in;
    program cdparanoia not setting count and/or reply_len properly

Despite these warnings, track ripping appears entirely normal.

---------------
Behavior or cdparanoia-III-10.2.i386-linux-elf -A (static version from
xiph website) executed as normal user or as root. Example below shows
results obtained running as root, from console window:

    root@nu: ./cdparanoia-III-10.2.i386-linux-elf -A
    cdparanoia III release 10.2 (September 11, 2008)

    Using cdda library version: 10.2
    Using paranoia library version: 10.2
    Checking /dev/cdrom for cdrom...
            Testing /dev/cdrom for SCSI/MMC interface
                    SG_IO device: /dev/scd0


[That's it.  After about 30 seconds, death:  Symptoms: Hard freeze, cursor
stops blinking, no keyboard response, no indication of disc or Ethernet
activity, no panic messages, no nothing.  After reboot, system logs contain
no hint of what might have happened.]

---------------
A few random observations:

  * System basic info: Thinkpad T43, CentOS 4-1.2, upgraded to 2.6.17 kernel.

  * CD drive info as reported by 9.8 above.

  * During the 30-second pre-death period (after the "SG_IO device: /dev/scd0"
    message appears, but before the freeze) I'm unable to ^C out of cdparanoia,
    but system operation in other respects generally appears normal.  For
    example, I had enough time to cut-n-paste the above pre-crash text into
    another console window and write it to a file before it froze.

  * My locally-built version of 9.8 (which I've been using for years, built
    sources in 2003 I think) also works fine, but does not seem to generate
    the "not setting count..." warnings.  I only see those messages with the
    9.8 static version from the xiph website.
----------------
_______________________________________________
Paranoia mailing list
Paranoia@...
http://lists.xiph.org/mailman/listinfo/paranoia

[xiphmont]

> I recently attempted to upgrade from cdparanoia 9.8 (which I've been using
> successfully for years on a variety of systems) to 10.2.  Alas, 10.2 seems
> to be causing hard, 100% reproducible crashes during drive detection, even
> running the static binary from the xiph website.
>
> I've done my homework on the website (bug reports, "troubleshooting" section,
> mailing list) and googled of course, but have not seen any mention of this.

Ah, tripping a bug in the kernel is never fun to debug.  I have a few
guesses as to what may be going wrong, and the kernel having an
overinflated idea of what DMA transfer sizes the Thinkpad can handle
is high on the list.

I'm going to be doing a sweep of the various bug reports that have
come in since 10.2 for an upcoming release soon and will have a few
things for you to try if you're still there and still willing.

Thanks for the report!  Reports are important!

Monty
_______________________________________________
Paranoia mailing list
Paranoia@...
http://lists.xiph.org/mailman/listinfo/paranoia

[Glenn Golden]

Hi Monty,

Any update on this? Anyone else reported anyting like it?

I'm happy to help if I can. Let me know what you'd like me to try.

Thanks,

Glenn


xiphmont@... writes:

> > I recently attempted to upgrade from cdparanoia 9.8 (which I've been using
> > successfully for years on a variety of systems) to 10.2.  Alas, 10.2 seems
> > to be causing hard, 100% reproducible crashes during drive detection, even
> > running the static binary from the xiph website.
> >
> > I've done my homework on the website (bug reports, "troubleshooting" section,
> > mailing list) and googled of course, but have not seen any mention of this.
>
> Ah, tripping a bug in the kernel is never fun to debug.  I have a few
> guesses as to what may be going wrong, and the kernel having an
> overinflated idea of what DMA transfer sizes the Thinkpad can handle
> is high on the list.
>
> I'm going to be doing a sweep of the various bug reports that have
> come in since 10.2 for an upcoming release soon and will have a few
> things for you to try if you're still there and still willing.
>
> Thanks for the report!  Reports are important!
>
> Monty

_______________________________________________
Paranoia mailing list
Paranoia@...
http://lists.xiph.org/mailman/listinfo/paranoia