OK, here is the complete output of cd-drive:

pi@autoradio:~ $ cd-drive
cd-drive version 0.92 armv7l-unknown-linux-gnueabihf
Copyright (c) 2003-2005, 2007-2008, 2011-2013 R. Bernstein
This is free software; see the source for copying conditions.
There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE.
The driver selected is GNU/Linux
The default device for this driver is /dev/cdrom

Drivers available...
  GNU/Linux ioctl and MMC driver    
  cdrdao (TOC) disk image driver    
  bin/cuesheet disk image driver    
  Nero NRG disk image driver        

CD-ROM drive supports MMC 3

                       Drive: /dev/cdrom
Vendor                      : MATSHITA
Model                       : CD-RW  CW-8124 
Revision                    : DA0D
Profile List Feature
        Read only DVD
        Read only Compact Disc capable - on
        Write once Compact Disc capable

Core Feature
        ATAPI interface

Morphing Feature
        Operational Change Request/Notification not supported
        Synchronous GET EVENT/STATUS NOTIFICATION supported

Removable Medium Feature
        Caddy/Slot type loading mechanism
        can eject the medium or magazine via the normal START/STOP command
        can be locked into the Logical Unit

Random Readable Feature

Multi-Read Feature

CD Read Feature
        C2 Error pointers are supported
        CD-Text is supported

DVD Read Feature

Incremental Streaming Writable Feature

Formattable Feature

Restricted Overwrite Feature

CD Track at Once Feature

CD Mastering (Session at Once) Feature

Initiator- and Device-directed Power Management Feature

CD Audio External Play Feature
        SCAN command is supported
        audio channels can be muted separately
        audio channels can have separate volume levels
        256 volume levels can be set

Ability to respond to all commands within a specific time Feature

Ability to perform DVD CSS/CPPM authentication via RPC Feature
        CSS version 1

Ability to read and write using Initiator requested performance parameters Feature

Vendor-specific code 10b Feature

Hardware                                  : CD-ROM or DVD
Can eject                                 : Yes
Can close tray                            : Yes
Can disable manual eject                  : Yes
Can select juke-box disc                  : No

Can set drive speed                       : No
Can read multiple sessions (e.g. PhotoCD) : Yes
Can hard reset device                     : Yes

Reading....
  Can read Mode 2 Form 1                  : Yes
  Can read Mode 2 Form 2                  : Yes
  Can read (S)VCD (i.e. Mode 2 Form 1/2)  : Yes
  Can read C2 Errors                      : Yes
  Can read IRSC                           : Yes
  Can read Media Channel Number (or UPC)  : Yes
  Can play audio                          : Yes
  Can read CD-DA                          : Yes
  Can read CD-R                           : Yes
  Can read CD-RW                          : Yes
  Can read DVD-ROM                        : Yes

Writing....
  Can write CD-RW                         : Yes
  Can write DVD-R                         : No
  Can write DVD-RAM                       : No
  Can write DVD-RW                        : No
  Can write DVD+RW                        : No

The problem with the ioctl play commands (there are a handful of them) is that they don't tell the drive how to deliver the (digital) audio data to the host computer: Old drives had their own decoders and sound card links, so that one single command was absolutely sufficient. Nowadays, a more complicated solution like cd-paranoia is necessary: cd-paranoia writes the audio data into a WAV file. If you replace the output file with a fifo and use the fifo as an input for, say, aplay, you'll have an audio CD player in software.

As someone who's run into exactly the same problem (on a Raspi 3B) and at least has a clue about its cause, may I hereby join this discussion:

Apparently, CDIO is using the ioctl commands CDROMPLAYMSF. See lib/driver/gnu_linux.c, line 262. This command will only work for drives with on-board audio decoder hardware and a corresponding audio output connector, which is then attached to the sound card. Nowadays, CD-ROM drives / CD recorders are no longer equipped with such hardware and simply ignore the ioctl command (without even issuing an error message).

This means: audio_play_msf_linux () won't work anymore. The only workaround I see rising above the horizon is to use cd-paranoia and to redirect its output to a command-line player like aplay.

Any suggestions?

Do you use XBMC or any other software to play CD?

Hi, I am trying to get this work on Win32 but it doesn't seem to work for audio CDs on external USB player.

Any suggestions?

The bug in xbmc is fixed now for GNU/Linux x86 and RaspberryPi. Both of them use driver GNU/Linux, access mode: IOCTL. Nobody reported yet if the patch works on win32 also.
I think you could apply patch to libcdio to prevent similar problem in any other program using the library.

There is a problem with "cdio_read_audio_sectors" function. When it reads more than about 10 sectors at once with USB drive it returns an error. When XBMC is modified to read less than 10 sectors at once it reads audio CD with USB drive. I'm not sure if bug is in libcdio.

Get the libcdio source code. Look at examples/drives.c.
Compile it and run it with a CD-DA disk in the drive.

Save and send that output. Also giving some other evidence using some other command that indicates it finds a CD-ROM drive with a CD-DA disk.

That's the basic idea. But to be more helpful, in examples/drives.c change the subroutine log_hander() to print all debug and info messages. Also you can remove the last to C statements and from main() before the final "return 0" since those look for kinds of drives other than CD-DA which isn't necessary here..

Okay, I have lookeda thte XMBC error report. But I can't go with something as vague as:

    The word around is that libcdio is the culprit as it doesn't handle USB optical drives correctly (Audio CDs at least).

I suggest writing some short piece of code give more concrete evidence. For example compile examples/drives.c change log_hander to print all debug and info messages and probably removing the last to statement of main before return 0.