So way back in December, I said I'd be adding 4 logs, and then I only added 3. Well, here's the 4th, and 2 others will follow.
So, sometime (probably back in December) I finally decided I'd go ahead and attempt to add a PATA bus to the DU600 board. In a previous log, I mentioned how I upgraded the ISA bus signals to support 16 bit transfers, and this was the main reasoning behind this. As it turns out, a simple PATA bus is actually a subset of the ISA bus. The way it is designed, PATA requires 16 bit transfers to be supported. This is because the data register is a 16 bit register, and the other registers are all 8 bit. Well, the data register actually ends up overlapping with another register partially, because x86 I/O ports are byte addressable, and there's no gap to allow for the other 8 bits of the data register before you're accessing the next register.
So how does this work? The PATA device, when any register other than the data register is accessed, tells the ISA bus to use 8 bit transfers (on the low address lines, D0-D7) by leaving the /IOCS16 line unasserted. When the data register is addressed, the PATA device asserts the /IOCS16 line, telling the bus that this address is capable of a 16 bit transfer. Even though the data register overlaps another register by one byte, it is in fact uniquely addressed, so this works out fine. You can see the other log for a bit more information on the work I did to support these 16 bit transfers.
Aside from that, the ISA bus --> PATA bus adaptation is fairly simple. Many signals carry over 1:1. Data lines, address lines, control signals. The only signals you really need to generate are the two chip select signals which are normally decoded from the high order address lines. One chip select selects one set of registers, and the other selects another set. Additionally, a good PATA interface would buffer the data bus lines to avoid bus loading issues.
For my circuit, I got away with using only one chip, a 74ACT138 for decoding the chip selects. I found an I/O range that didn't conflict with the VGA chip, and decoded the two PATA chip selects in that range. I should really have a buffer for the data lines, such as a 74ACT245, but there don't seem to be any issues without it as long as the connected drives are actually powered. The rest of the wiring was, well, just wires.
Another gotcha here, again, is that the 68K is big endian, and the x86-based PATA bus is little endian. Doing the proper byte swap in hardware here with the data lines solves most issues, just like on the SVGA controller. This ensures the same byte ordering within a sector of the drive as a PC would produce, but with the caveat that 16 bit values must be byte swapped in software (same caveat applies to the SVGA controller, although the only 16 bit values that exist there are within the blitter functionality).
Once I had completed the hardware, I took a look at some of the various PATA tutorials online, and issued the ATAPI IDENTIFY command to a CDROM drive I had laying around (I didn't have any HDD's around when I finally got to doing test software). Although the software was crude, it worked and listed the hex values of all the bytes within the returned block from the IDENTIFY command. I was able to decode all of the info and see various parameters like drive model and serial number, which told me the hardware was working.
I haven't done much with the PATA hardware since, I've been focusing on Joe-Mon instead.