VirtualBox SoundBlaster 16 Issue

[SpyderTL]

I've run into a minor issue with the SoundBlaster 16 running on VirtualBox. Before I head over there and put in a bug report, I want to make sure that I'm not missing something...

According to the SB programmers guide, and several web sites that I've seen, the SB16 sends an interrupt (IRQ 5 on VirtualBox) when the audio buffer has finished playing, and it sets a flag in mixer register 82h. The driver/interrupt handler is supposed to acknowledge the interrupt by reading 1 byte from the base I/O Port + 0Fh (in 16-bit mode). So far, so good.

The problem is that the interrupt flag (bit 1) on register 82h is never cleared, even after the I/O port has been read. I was using this flag to determine when to re-fill the DMA buffer, but after it is set the first time, it is never cleared, so it always appears that an interrupt just occurred. I tried several things to clear this flag -- reading from 0Fh, writing to 0Fh, writing to register 82h, etc, but I can't find any way to clear the flag. I would assume the flag should be cleared after the 16-bit status port is read (0Fh), but the documentation doesn't mention this anywhere, and I can't find any references to this being a problem on google.

Would anyone happen to know if this is actual SB16 (DSP 4+) behavior, or if this is a bug in VirtualBox?

Right now, I'm just using the interrupt to fill the buffer, and ignoring the interrupt flag (but still sending the acknowledge by reaing 0Fh). But if any other device ever uses the same IRQ, this will become an issue real quick.

I may try digging through the VirtualBox source and see if I can figure out what is going on.

Thanks guys.

EDIT: After looking through the source, I found this:

Code: Select all

1164    case 0x0e:                  /* data available status | irq 8 ack */ 
1165        retval = (!s->out_data_len || s->highspeed) ? 0 : 0x80; 
1166        if (s->mixer_regs[0x82] & 1) { 
1167            ack = 1; 
1168            s->mixer_regs[0x82] &= 1;                 // PROBLEM!
1169#ifndef VBOX 
1170            qemu_irq_lower (s->pic[s->irq]); 
1171#else 
1172            PDMDevHlpISASetIrq(s->pDevIns, s->irq, 0); 
1173#endif 
1174        } 
1175        break; 
1176 
1177    case 0x0f:                  /* irq 16 ack */ 
1178        retval = 0xff; 
1179        if (s->mixer_regs[0x82] & 2) { 
1180            ack = 1; 
1181            s->mixer_regs[0x82] &= 2;              // PROBLEM!!
1182#ifndef VBOX 
1183            qemu_irq_lower (s->pic[s->irq]); 
1184#else 
1185            PDMDevHlpISASetIrq(s->pDevIns, s->irq, 0); 
1186#endif 
1187        } 
1188        break;

Shouldn't that be &= ~1 ??

Anyone able to submit this fix to the VBox repository, or should I just put in a bug report. Is there a way I can submit this bug fix without downloading the entire project?

Thanks guys.

EDIT: Submitted ticket: https://www.virtualbox.org/ticket/13769

[Geri]

i think you should not expect hardwares/emulators to behave exactly as the specifications say

i remember back then sharing the same irq for sound card with other device resulted immediately bsod on windows 9x on boot
(and this is true on almost all other device)

i suggest you to try to find out more esoteric ways, where the playback is - you know the elapsed time, the buffer size, so you know when it should run out

[Brynet-Inc]

SpyderTL's fixes were correct and incorporated into both VirtualBox and QEMU. The world rejoices as Sb16 emulation accuracy improves.