Subject: Re: CP-M Z80 home brew computer circuit board From: "Eric Smith" <eric at brouhaha.com> Date: Tue, 17 Jun 2008 11:53:23 -0700 (PDT) To: "General Discussion: On-Topic and Off-Topic Posts" <cctalk at classiccmp.org> Allison wrote: I think you also need some logic to release the ready line if DRQ doesn't occur but an interrupt does, in order to handle error cases.

It does, but on the NMOS parts it only can stall read cycles, not writes. (In the early-to-mid 1970s, ROMs and EPROMs were much slower than typical SRAMs, so the designers apparently didn't think there was any need to stall on writes.)

For the controller I designed, I dealt with that by having a special address that could be read that only waited for DRQ, at an address in the page below the FDC's data register. By itself, that wasn't sufficient to get to 8" MFM on a 1 MHz 6502. However, by taking advantage of the spurious read that occurs on indexed write instructions when there is a page crossing, I got the read of the special polling adress and the write of the data in a single instruction. That gets it down to 16 cycles, which meets the nominal rate but isn't actually good enough. By unrolling the loop, though, it drops to 13 cycles per byte, which meets all the requirements:

LDA (DPTR),Y ; get data byte from buffer STA FDC_SPCL,X ; reads poll location, which stalls, ; then writes data INY The unrolled loop requires 6 bytes of code per byte transferred, which is 1536 bytes for a 256-byte sector. Not great, but it did fit in a 2716 EPROM. I used two 2716s, one containing the read loop and one containing the write loop, both mapped into the same address space and selected by writing to another port address. The rest of the FDC driver fit in the remaining space of the two EPROMs. I didn't anticipate that they would "fix" the RDY behavior and the spurious read on the 65C02, so the same code wouldn't work on that. However, the fix was simply to replace the indexed STA with an absolute STA.

Eric