Bug 25346

The FreeBSD audio driver recognizes the CS4614 as csa0 on irq 9. The
driver works fine for audio playback and capture EXCEPT the the tail
of a playback will contain a bit of "buzz" after the real data bits
are output. This can be subdued by writing 4096 zero bytes at the end
of the data. This "buzz" has been traced to the fact that the csa0
interrupt service routine is never called. Vmstat -i also shows no indication
that anything talks to irq 9. The irq line can be changed via BIOS setup
but always audio and graphics share the same line.

In an attempt to debug this, I disabled the integrated audio chip and installed
a Turtle Beach Santa Cruz PCI card. This card has a CS4630 chip which is
compatible with the CS4614. The Dell motherboard has four 32-bit PCI slots
and four 64-bit PCI slots. One of the 64-bit slots can take a 32-bit card.
The graphics card is in the first 32bit slot. All other slots are empty.

I found that by installing the sound card in either the last 32-bit slot or
the shared slot, vmstat -i would show a "mux" device at the irq of the
sound card and interrupts would be delivered to the sound driver. Not
so if the sound card was installed in slots 2 or 3. In slot 4, the
sound device shared an interrupt with one of the AHC SCSI devices. In
slot 5 (the shared 32/64 bit slot), the sound device shared an irq
with two other devices.

I also have a Dell Precision Workstation 420 available which also has the
same integrated audio chip. There is no problem with interrupt delivery
on that system. The bus structure as reported by dmesg is also different.

How-To-Repeat: 
Dmesg shows the csa0 and pcm0 devices should be using irq9. Running
"systat -vmstat 1" while playing or recording using the sound device
shows no interrupt activity on irq 9 or what ever line is assigned
in BIOS setup unless an non-integrated sound card is plugged into
pci slots 4 or 5.