Sound and other 16-bit DMA drivers should write -1 to length on completion

[TimeHorse]

Based on ​this forum post, the drivers using Direct Memory Addressing should write -1 (or 0xFFFF / 65535) to the length field of the buffer once the DMA transfer is complete. I know this was an issue with the version unspecified in the post and appears to be an issue in the current version from my observations. However, I've yet to investigate the code to verify it behaves as I describe (neglecting to set -1 to indicate completion). If it now does that, I apologize for wasting everyone's time as the current sound driver issue in DOC must have anther origin. If it doesn't, I would appreciate a hint as to where the DMA code lies to see if I can fix it myself.

Thanks.

Steps to reproduce:

1. Install a DOS guest OS
2. Install the Soundblaster driver and set the BLASTER variable to "A220 I5 D1 T6 H7 P330"
3. Install USound.exe from the Ultima Underworld installation or other source.
4. Run USound.exe; no sound card detected despite setting as specified above (0x220 Addr, IRQ 5 DMA 1, etc.)

[aeichner]

Please reopen if still relevant with a recent VirtualBox release.