zipcpu / qspiflash

A set of Wishbone Controlled SPI Flash Controllers

The Zip CPU

The Zip CPU is a small, light-weight, RISC CPU. Specific design goals include:

• 32-bit. All registers, addresses, and instructions are 32-bits in length.

• A RISC CPU. Instructions nominally complete in one cycle each, with exceptions for multiplies, divides, memory accesses, and (eventually) floating point instructions.

  (Note that the ZipCPU is not a RISC-V CPU, nor does it copy from any other instruction set but its own.)

• A load/store architecture. Only load and store instructions may access memory.

• Includes Wishbone, AXI4-Lite, and AXI4 memory options.

• A (minimally) Von-Neumann architecture

  The Wishbone wrappers share buses for instructions and data, the AXI4-Lite and AXI4 wrappers have separate bus interfaces for each. The address space itself, however, needs to be common.

• A pipelined architecture, having stages for prefetch, decode, read-operand(s), a combined stage containing the ALU, memory, divide, and floating point units, and then the final write-back stage.

• A two mode machine: supervisor and user, with each mode having a different access level.

• Completely open source, licensed under the GPLv3.

Unique features and characteristics

• Only 29 instructions are currently implemented. Six additional instructions have been reserved for a floating point unit, but such a unit has yet to be implemented.

• (Almost) all instructions can be executed conditionally. Exceptions include load immediate (LDI), the debug break instruction (BREAK), the bus lock (LOCK) and simulation instructions (SIM), and the no-operation instruction (NOOP). The assembler will quietly turn a conditional load immediate into a two-instruction equivalent (BREV, LDILO).

• The CPU makes heavy use of pipelining wherever and whenever it can. Hence, when using a pipelined memory core, loading two vaues in a row may cost only one clock more than just loading the one value.

• The CPU has no interrupt vectors, but rather two register sets. On any interrupt, the CPU just switches from the user register set to the supervisor register set. This simplifies interrupt handling, since the CPU automatically saves, preserves, and restores the supervisor's context between enabling interrupts and receiving the next interrupt. An optional interrupt peripheral may be used to combine multiple external interrupts into a single interrupt line.

Getting Started

If you'd like to get started with the ZipCPU, you might wish to know that this repository contains the CPU, its documentation, and the toolchain. The CPU implementation found here, though, is just that: a CPU. This implementation requires a bus with peripherals hanging off of it, things such as RAM, flash (ROM), serial port, etc. This is just where I keep the CPU apart from any necessary peripherals.

So, if you want to try out the CPU, feel free to download and build this repository (use git-clone with a depth of 1--there's a lot of stuff in the git repo that you don't necessarily need). You'll need it for the binutils, GCC, and newlib support provided by it.

Once you've built these tools, then I'd suggest you look into the ZBasic repository. That repository places the CPU in an environment with block RAM, QSPI flash, and SD-card (SPI protocol) access. From that repository, you can either tweak the distro (main.v, regdefs.h, board.h, board.ld) to add the peripherals you want to use the CPU with, or you can use AutoFPGA to adjust your RAM size, add or remove peripherals and so forth while maintaining (creating, really) all of these files for you.

The sim/ subdirectory also contains a version of the ZipCPU in a usable environment for simulation purposes. This includes the CPU, possibly more CPU's for a multiprocessor environment, bus interconnect, memory, a simulated serial port, and a couple more peripherals.

If you aren't interested in simulating the CPU, there is an assembly level debugger that you can use to stop and step the CPU, as well as an integrated wishbone scope that you can use to get traces from within the design while it is running.

Need help?

If you'd like to use the ZipCPU, and don't know where to begin, feel free to find me on IRC as ZipCPU. I've created a #zipcpu channel on several IRC servers that I tend to inhabit. If you get stuck, feel free to drop by and ask for help. You can also drop by just to say hi. Either way, please introduce yourself and give me some time to respond.

Current Status

You can also read about the ZipCPU via several blog articles posted at zipcpu.com! Articles you might find valuable include:

• An overview of the ZipCPU's ISA
• How-to build the tool-chain, and test the CPU
• Formal properties of a WB bus
• Formally proving the prefetch

Not yet integrated

• The MMU that was written for the ZipCPU now needs to be rewritten in order to work in the new ZipCore context. My plan is to place the MMU between the ZipCore and the various bus aware wrappers, providing a similar interface to the one offered by the ZipCore. That way, caches will be based on physical addressing--solving one of the bigger problems I had when using the MMU.

• FATFS support now exists for the SDCard, it's just not (yet) integrated into the newlib C-library.

• The ZipOS would greatly speed up and improve the bare bones newlib library--primarily by providing "O/S" level interrupt support when using the library. This integration has not (yet) been accomplished.

  On the other hand, if the MMU is available, I might rather just create a Linux distribution.

Commercial Opportunities

The GPLv3 license should be sufficient for most (all) academic and hobby purposes, and certainly for all simulation based purposes. If you find, however, that the GPLv3 license is insufficient for your needs, other licenses can be purchased from Gisselquist Technology, LLC.

This applies to all but the toolchain, for which GPLv3 should work for all purposes. (Besides, I don't own the licenses for Binutils, GCC, or newlib.)

qspiflash's Issues

Configuration read doesn't work, returns 0x00 or 0xFF

I managed to add dualflexpress.v to my branch of the CPU86 project (a free 8088 CPU implementation design), running on the MAX1000 board. I wrote a little glue logic to map an MMIO interface for the flash controller. Reading from the flash controller works with that now.

Configuration writes also appear to work -- at least, after calling take_offline and then place_online in my check.asm then the read tests work again (same as right after programming the FPGA). Those are c86boot.asm -D_TESTLOAD=1 and test6.asm -D_AMOUNT=20 -D_ADDRESS=0 -D_STRIDE=1.

Configuration reads however seem to return either 0x00 or 0xFF, not the data they should be reading. With the current check.asm file I get "00000000" for the flash ID and twice "00" then 14 times "FF" for the read test. (Same results with the QSPI controller instead, which I replaced by the DSPI controller.) (Tried with check.asm's define -D_EMPTY_BEFORE_READ from 0 to 10.)

Did I get something wrong?

qflexpress.v on MAX1000 board (W74M64 Winbond flash) fails to read, always gives 0xCC or 0xEE

Just today I got dualflexpress.v to work on my MAX1000 board (W74M64 Winbond flash of 8 MiB, written to using max1000_flashrom). I had to replace localparam by parameter to make Quartus happy. I also deleted the assign to o_dbg_trigger and o_debug, and commented out these two output wire definitions. I set CLKDIV to 1, RDDELAY to 0, and NDUMMY to 4. I uploaded the project with the successful tweaks to https://hg.pushbx.org/ecm/c86dled/rev/d787dedd0692 (It is a very simple design, using the 12 MHz clock, the user input button, and the 8 user LEDs to output the data read byte by byte. (Every 5th output is the same as the 4th.))

qflexpress.v doesn't work however. To make it compile I modified it in the following ways: parameter instead of localparam, DW=32 had a trailing comma it shouldn't have, and I tried CLKDIV = 1, RDDELAY = 0, NDUMMY = 0. This combination leads to reading always 0xEE (1110_1110b bytes). Other combinations lead to reading always 0xCC (1100_1100b bytes). Unlike my experience with dualflexpress, qflexpress never seems to return any of my data, not even partially / corrupted. I uploaded this design to https://hg.pushbx.org/ecm/c86qled/rev/6991862071ba (It is largely the same as the first one, just using qflexpress instead of dualflexpress.)