mainframe.dev

The last days have been filled with a lot of Verilog and ModelSim, so I thought it would be nice with some softer programing for a change. After spending more time than I care to admit on hunting down the architecture reference guide (it is here if you need it) I figured I could as well write the bootloader that my OpenRISC platform will require. Since I haven't written anything about what I have in mind, here is a quick overview what I'm looking at creating: Implement OpenRISC Make the platform fit the DE-2 board from Terasic It must be able to boot Linux It would be awesome if I can get the DM9000 network controller working That is the very formal (hmm) requirement specification I have in mind. Anyhow, to boot Linux on the DE-2 platform I would need to have a big RAM (hence the SDRAM controller) of a couple megabytes at least. I would also need to have some permanent storage to boot from; the flash memory will be good for that I think. When booting ...

I think I pulled this one off quite nicely. The following is a controller for the 8 Megabyte SDRAM that the DE2 board uses. Features: Basic command queue ("run this when possible") Read / write support Automatic refresh Gotchas: Uses burst of only one word, it is in other words quite slow Hardcoded to work with 50 MHz and CAS 2 The controller is very very simple and always executes the read and write operations as pre-charge variants which will close the activated row after the operation. The result of this is that a single read of 2 bytes will take 5 clock cycles (activate, read, CAS 2, actual read) - a write will take 3 (activate, write, recovery). I'm inclined to create something more intelligent, faster and using the other features that this SDRAM module has to offer - but this will do for now.

I read a lot of blogs about people doing hardware work just as I like to do. A long time ago a person who I have the utmost respect for failed to create a SDRAM (or DRAM, I can't really recall) controller which made me sure that I wouldn't be able to do it. In order to continue with OpenRISC I will need some sort of big RAM, and after a slight nudge from a friend of mine I began reading about the SDRAM my DE2 board ships with. Two days of tinkering produced this ugly (but working!) FSM. It is locked to using row 0x0 but writes a pattern to each column address (there are 256 of them on this SDRAM). After the write it will jump to a read / refresh loop which basically proves that it is capable of holding on the data. Pretty neat, this will make a great start for a SDRAM controller.