Hi All!
TLDR
I was able to get DMA and ISA sound cards working on this motherboard. Version 3.3 of this motherboard has a design defect where the PPDREQ# and PPDGNT# lines from the ITE8888-F to the ICH4 south bridge are swapped. Fixing this is possible, but not trivial. It involves cutting the traces involved and swapping the signals somewhere on the board. If you have one of these and you want to try it, of course you do so at your own risk. I don't know that versions other then 3.3 have this issue (or even exist). Don't do this if you're having any other problems (like PCI card sound problems) - this won't help with that.
The Symptom
ISA sound cards in this board play music only, but there is no "digital audio". Most utilities or games won't complain about it unless they actually test DMA. An example of such a utility is the creative diagnose.exe program which, when run will have no issues setting the base at, MIDI, and IRQ at standard locations. When the DMA test is run, it'll return that all DMA lines seem used and you can't get past the test. Note that the BIOS reserves DMA 3 for the parallel port and it seems to do this even if the port isn't set to EPP mode. It could reserve a different DMA for the parallel port if you select a different one in the CMOS setup. I didn't check, but I strongly suspect it would release this if you disable the parallel port completely.
The Problem
The ICH4 supports a total of 6 PCI bus masters for "external" PCI devices on the bus. There are 4 dedicated Request/Grant (REQ/GNT) pairs for this REQ(0:4) and GNT(0:4), one dedicated pair for PC/PCI DMA (PPDMA) REQ(A)/GNT(A) and one multiplexed pair which can be used for either function REQ(5/B) and GNT(5/B). The pins for REQ(A)/GNT(A) and REQ(B)/GNT(B) are triple-duty and can also be used as GPIO, but this board doesn't use them for that.
Because this board has 2 on-board 1GE ethernet interfaces and 4 PCI slots, all the available REQ/GNT pairs 0-5 are used and REQ(B)/GNT(B) is unavailable for PPDMA. The ITE8888F is using REQ(A)/GNT(A) and due to the lack of available pairs on the ICH4 its IGNT# pin is pulled high with a 4.7k pull-up and IREQ# pin is unconnected. This means that the ITE8888F can not act as a PCI BUS master and any attempt to use distributed DMA (DDMA) on this board won't work (without modification). You can get more information in that in the linked thread. Thanks so much to the O/P KYA for helping me work through some of this! AWE32 and PicoGUS on a Core2Duo industrial motherboard
The Diagnosis
In the thread linked, a great deal of time was spent vetting the BIOS' work setting the registers in the ICH4 and the ITE8888F only to find no issues with how the BIOS was configuring things. After being frustrated with this and worried I missed something I finally bit the bullet and decided I'd attempt to trace out the two signals critical to PPDMA from the ITE8888 to the ICH4. To see the traces clearly, I had to remove one electrolytic cap, the battery holder, a ceramic cap and a resistor network near the bridge. Even then, since the ICH4 was a BGA part and I wasn't about to pull it off and attempt to put it back on there was still the risk that if a trace went under it on the top layer I may not be able to see where it went. At this point I was 99.9% sure the PPDMA REQ/GNT pair had to be using REQ(A) and GNT(A) of the ICH4 because nothing else made sense hardware-wise and the GEN_CNTL register in the ICH4 was configured to make REQ(B)/GNT(B) be used as REQ(5)/GNT(5).
If you check for continuity and you get continuity as per the image below - the signals are swapped. Pin numbers are in reference to the signals pin on the ITE8888F.
While it's unclear from the image exactly where on the ICH4 PPGNT# ends up, we can determine with certainty on the back, that PPREQ# ends at the ball for GNT(A), and based on the relative location where the trace goes under the ICH4, the fact that REQ(A) is ball B5 near the outside of the chip and the fact the bottom photo confirms GNT(A) is connected it's a pretty safe bet that PPGNT# is ending at REQ(A). Could it be going to REQ(B) at ball A6? Yes, it could be, but had they done that at least one of the PCI Slots or one of the NICs wouldn't have worked as a bus master. Now their quality control here is questionable, but that would be harder to overlook then just DMA not working on the ISA slots.
The Fix
In theory of course the fix is easy - just swap the signals. In practice, maybe not so easy because it means you need to cut the traces to break the original signals and solder a jumper to swap them. Since I spent several hours tracing them both I had the opportunity to find where on the board I thought would be the easiest place to swap them. That particular spot is on the back of the board under where the battery holder is. Here the two signals run parallel to each other through two via's from the top, down to the bottom for a short time, then back to the top toward the ICH4. If you fix this yourself you may find and easier spot to use, this is just what I did. At any rate, mind the length of the wires since if my DDMA experiment from the other thread proved anything it showed that having two really long bodge wires of radically different lengths routing these signals from one side of the board to the other could cause bus timing problems....
And finally, the traces swapped. And yes yes I know I'm not perfect and I did clean up where I slipped there cutting the traces and I put some more solder mask in places where I exposed the copper and all that jazz..
If you do it this way ensure your bodge wires aren't shorted to each other and that the traces were cut cleanly. re-check continuity back on the resistor networks to ensure the signals were swapped and you should be done (in much less time then it took me to figure this out).
PPDMA on the PCI-to-ISA bridge should work now, and you should get sound from your ISA sound card. Good luck and I hope this helps someone with this board, or saves someone some money in not buying a board with a design defect!
