More "already active DMA on this device"

Hi, everyone.

I seem to have a case of the "already active DMA" problem on recent -CURRENTs 
on my laptop.  So far what I've found in the archives has suggested ich5 
controllers or apic changes as possible causes, but I have neither of those.  
Here's a quick summary possibly relevant info:

* No apic or ACPI on this system (Compaq m700 -- it does have ACPI capability 
but is so horribly broken that I use apm instead)
* Controller is a CMD646U2.  Yeah, I know, but it's embedded in a docking 
station so I'm kind of stuck.
* Controller is behind a PCI bridge (Intel 82380FB)
* Pretty much EVERYTHING in the system is sharing irq 11.  I've tried various 
methods to coax devices onto other IRQs, but have had no luck whatsoever.  It 
seems to be a BIOS or maybe even a hardware limitation.
* The fxp0 controller behind the same brdige seems to work OK
* The firewire (Lucent FW322) controller behind the bridge is flaky, even with 
kernels where the ATA works.  By flaky, I mean works for a while then freezes 
up (no more interrupts maybe?)
* A cardbus firewire controller (VIA VT6306) works fine, so I can do 
postmortem memory investigation using that if necessary.

Messages when it starts:
ad6: FAILURE - already active DMA on this device
ad6: setting up DMA failed
ad6: TIMEOUT - READ_DMA retrying (2 retries left) LBA=153861628
ata3: reiniting channel ..
ata3: reset tp1 mask=03 ostat0=58 ostat1=08
ad6: stat=0x50 err=0x01 lsb=0x00 msb=0x00
ata3-slave:  stat=0x00 err=0x01 lsb=0x00 msb=0x00
ata3: reset tp2 stat0=50 stat1=00 devices=0x1<ATA_MASTER>
ata3: resetting done ..
ad6: pio=0x0c wdma=0x22 udma=0x45 cable=80pin
ad6: setting PIO4 on CMD 646U2 chip
ad6: setting UDMA33 on CMD 646U2 chip
ad6: setting UDMA33 on CMD 646U2 chip
ata3: device config done ..

I usually have /usr and other important partitions mounted from this drive, so 
a UFS panic is typically not far behind.

It seems to still be present in very recent CURRENT too.  So far I've found:

5/7/2004	Works
6/15/2004	Broken
7/2/2004	Broken

Since it doesn't present itself immedately but rather a while after boot, I 
haven't been able to narrow it down to a particular commit.  Verbose dmesg is 
attached (from 6/15 kernel).  If there's any more information I can post or 
debugging I can try that might help, I'd be more than happy to give it a 
shot.

Thanks,
Craig