Subversion Repositories s6soc

0.2 & 5/14/2016 & Gisselquist & Updated Draft, still not complete \\\hline

the CPU.  The CPU will then begin following the instructions found in flash

memory.

\chapter{Software}

\section{Directory Structure}

\begin{itemize}

\item[{\tt trunk/bench}] Contains software for emulating the S6 without the S6

        present.

  \begin{itemize}

        \item[{\tt trunk/bench/cpp}]  All of the bench testing software is

                written in C++, so it is found in this directory.  Primary

                among these programs is the {\tt zip\_sim} program which will

                simulate the ZipCPU within the S6--Soc.  Specifically, it

                simulates everything at or below the {\tt busmaster.v} level.

                Some, although not all, of the peripherals have been simulated

                and made a part of this simulation.  These include the

                Quad--SPI flash, the UART, and the LED's.

  \end{itemize}

\item[{\tt trunk/doc}] All of the documentation for the S6SoC project may be

        found in this documentation directory.  Specifically, I would commend

        your attention to anything with a {\tt .pdf} extension, as these

        are the completed documents.  Among these you should find a copy of the

        GPL copyright under which this software is released, as well as a

        pre--built copy of this document.

  \begin{itemize}

        \item[{\tt trunk/doc/gfx}] Here is where the graphics are located in

                support of this specification document.

        \item[{\tt trunk/doc/src}] And here is where the \LaTeX files are

                kept that were used in building both this document as well as

                the GPL copyright.

  \end{itemize}

\item[{\tt trunk/rtl}] Verilog files

  \begin{itemize}

        \item[{\tt trunk/rtl/cpu}]      Verilog files containing the ZipCPU

                core and peripherals.  The toplevel file here is the

                {\tt zipbones.v} file, although some of the peripherals, such

                as the {\tt ziptimer.v} are referenced independently.

  \end{itemize}

\item[{\tt trunk/sw}] The main software directory, primarily a repository

        for software subdirectories beneath it.

  \begin{itemize}

        \item[{\tt trunk/sw/dev}]  This directory holds a variety of

                simple programs for the ZipCPU, such as {\tt helloworld},

                {\tt doorbell} and {\tt doorbell2}, as well as software drivers

                for various peripherals, such as the real--time clock simulator,

                and the keypad and display device drivers.

        \item[{\tt trunk/sw/host}]  This directory holds support software which

                can be built on and run on the host machine.  Building this

                software will involve adjusting the Makefile so that it knows

                where your local ADEPT installation directory is.\footnote{Many

                of the programs also depend upon the serial number of my CMod

                S6 device.  This will need to be adjusted in any new install.}

                Once built, you will find a variety of very useful programs

                within here.

        \item[{\tt trunk/sw/zipos}]  This directory contains the source code for

                        a rudimentary, very basic, operating system that I

                        call the ZipOS.

  \end{itemize}

\end{itemize}

\section{ZipCPU Tool Chain}

To build programs for the ZipCPU, you will need the ZipCPU toolchain.  You

can find this as part of the ZipCPU project, available at OpenCores.  Building

the ZipCPU project should result in a set of binaries in the

\hbox{\tt trunk/sw/install/cross-tools/bin} directory.  Make this directory

a part of your path, and you should be able to build the CMod S6 ZipCPU

software.  Specifically, you will need to use {\tt zip-gcc}, {\tt zip-as}

{\tt zip-ld}, and {\tt zip-cpp}.  Other tools, such as {\tt zip-objdump} and

{\tt zip-readelf}, may also prove to be very useful when trying to figure out

what is going on within the SoC.

\section{Bench Test Software}

Bench testing software currently consists of the {\tt zip\_sim} program found

within {\tt trunk/bench/cpp}.  This program requires Verilator to run, and

simulates in a cycle accurate fashion, the entire S6SoC from {\tt busmaster.v}

on down.  Further, the external Quad--SPI flash memory, UART, and LED's are

also simulated, although the 2--line display, audio, and keypad are not.

\section{Host Software}

These include:

\begin{itemize}

\item {\tt dumpuart}: My current approach to debugging involves

        dumping the state of the registers and memory to the

        UART upon reboot.  The dumpuart command found here is

        designed to make certain that the UART is first set

        up correctly at 9600~Baud, and that second everything

        read from the UART is directly sent to both a file and

        to the screen.  In this fashion, it is similar to the

        UNIX {\tt tee} program, save for its serial port

        attachment.

\item {\tt readflash}: As I am loathe to remove anything from

        a device that came factory installed, the

        {\tt readflash} program reads the original installed

        configuration from the flash and dumps it to a file.

\item {\tt wbregs}

\item {\tt zipload}: This is the primary program you will need

        to get your software loaded on the CMod.  It takes three

        arguments.  The first is the name of the primary

        Xilinx configuration file, the second is the name

        of the alternate Xilinx configuration file, and the

        third is the name of the ZipCPU program you wish to

        write to Flash memory.

        Each of these arguments is optional.  For example, if

        only one configuration file is given, the loader will

        load the primary configuration.  If only one ZipCPU

        program is given, the program will be loaded into

        the program memory area and the configuration file areas

        will be left untouched.

\end{itemize}

\section{ZipCPU Programs}

\begin{itemize}

\item {\tt helloworld}: The first program any programmer should build,

        ``Hello, world!''  This program sends the string, ``Hello, world!''

        over the UART connection once per second.  It is a very valuable

        program because, if you can get this program running, you know you have

        a lot of things working and working correctly.  For example, running

        this program means you can run the {\tt zip-gcc} compiler, load

        the auxiliar configuration, load the program info flash memory, load

        the primary configuration, and read from the UART port.  It also means

        that you must have the UART port properly configured and wired to your

        CMod board.

\item {\tt doorbell}: This annoying program verifies the functionality of the

        audio device by playing a doorbell sound to the audio port.  It will

        then wait ten seconds, and play the doorbell sound again (and again,

        and again).  (It gets old after a while ...)

\item {\tt doorbell2}: This adds to the functionality of the {\tt doorbell}

        program a wait for keypress, and a display of the current time on the

        2--line display.  While almost our fully functional program, this

        does not include any menus to configure the device or set time, since

        it doesn't include any keypad functionality.

\item {\tt kptest}: A test of whether or not they keypad driver works.  When

        run, anytime a key is pressed, the key's value (in hex) will be

        sent to the UART.  Further, pressing an `F' on the keypad will also

        send a newline over the UART, in case you wish to keep your lines from

        getting too long.

\end{itemize}

\section{ZipOS}

This operating system is pre--emptive and multitasking, although with many

limitations.  Those familiar with the internals of the Linux kernel may laugh

that I call this an Operating System at all: it has no memory management unit,

no paging, no virtual memory, no file I/O access, no network stack, no ability

to dynamically add or remove tasks, indeed it hardly has any of the things

most often associated with an Operating System.  It does, however, handle

interrupts, support multiple pre--emptive tasks in a multitasking, timesharing

fashion, and it supports some very basic and rudimentary system calls.  In a

similar fashion, it does contain just about all of the functionality necessary

for a multi--tasking microcontroller built around a do--forever loop.  For its

size, I consider it an impressive achievement.  You are welcome to disagree

with me, however.

This version of the ZipOS starts in the resetdump.s code, so that upon

any startup the ZipOS will dump register contents, the BusError register, and

any scope contents to the UART.  This can take some time, so you may wish to

configure what you really wish to send--if anything.  If desired, this will

also dump the entire memory as well.  All of this is quite useful in case the

ZipCPU encounters a bus error or other sort of error that causes it to hang,

stall, or reboot, as these registers are very carefully not touched prior to

being sent to the UART output port.

{\tt resetdump.s} also calls a rudimentary bootloader, to load the parts of

the ZipOS that need to run faster into Block RAM.  The choice of what parts

to load into Block RAM is made on a file by file basis, and found within

the linker script, {\tt cmodram.ld}.

Upon completion, {\tt resetdump.s} calls the entry routine for the O/S,

{\tt kernel\_entry()} found in {\tt kernel.c}.  This is the main task loop for

the entire O/S, and worth studying if you are interested in understanding how

the O/S works.

The user tasks are found (mostly) within {\tt doorbell.c}, also found in the

ZipOS directory.  This file contains two kernel entry points, {\tt kntasks()},

which returns the number of tasks the kernel needs to know about, and

{\tt kinit()}, which builds each of the tasks and connects their file

descriptors to the various devices they will be referencing.

\subsection{Traps}

The ZipCPU supports a variety of traps, listed here:

\begin{itemize}

\item WAIT: Halts the execution of a process until an event takes place, or

        a timeout has been reached.  The events that can take place are a

        bitmask of the various interrupts the CPU supports, together with a

        bitmask of the values found in {\tt swint.h}.

        The timeout value can either be zero, to return immediately with the

        list of events that have taken place, negative, to wait indefinitely,

        or a positive number of milliseconds in order to wait at least that

        time for the event of interest to take place.

        This also allows a process to sleep for any number of milliseconds.

        When wait returns, any events returned by the wait have been cleared.

        The other thing to be aware of is that events may accumulate before the

        wait system call.  They will only be returned and cleared, though, if

        the wait call indicates an interest in those events.

\item CLEAR: This system call works closely with the wait system call.

        Indeed, it is very similar to a wait system call with a zero timeout.

        It clears any of the requested events which may be pending.  If given

        a timeout (in milliseconds), it will start a timer and generate a

        {\tt SWINT\_TIMEOUT} event to be sent to this process when the timer

        completes.

\item POST: Certain devices, such as the real--time clock and the doorbell

        reader, need the ability of being able to post events to any listener

        within the O/S.  The POST system call allows them to POST events in

        this manner.

\item YIELD: This is simply a way of being nice to other processes.  This system

        call takes no arguments and simply asks the scheduler to schedule the

        next process.  It does not take this process off of the ready to run

        list, so the next process may be this one.  However, since the scheduler

        is a round--robin scheduler, it will only return to this process if

        nothing else is available to run.

\item READ: This is roughly the same system call as the POSIX read() system

        call.  It reads some number of memory addresses (words, not octets),

        to the given file descriptor.  If the memory requested is unavailable,

        the read will wait until it is available, possibly indefinitely.

\item WRITE: This is roughly the same system call as the POSIX write() system

        call.  It writes some number of memory addresses (words, not octets),

        to the given file descriptor.  If nothing is reading from the device,

        the write stall the task forever, otherwise it will only stall the task

        until the data is either written to the receiving task, or copied into

        a memory buffer.

\item TIME: Returns the number of seconds since startup.  Eventually, this will

        return the number of seconds since January 1, 1970, and be identical

        to the UNIX system time() command, but that may not happen on this

        project.

% \item SEMGET

% \item SEMPUT

\item MALLOC: Allocates memory from the system/kernel heap.  This is a very

        low overhead memory allocator that, while it does allocate memory,

        cannot free it later.  It is nearly 100\% efficient since only

        one memory address, the top of the heap, is used to determine what

        memory has been allocated.

\item FREE: Not currently implemented.

\end{itemize}

\subsection{Scheduler}

The ZipCPU currently supports only a round--robin scheduler.  Tasks are executed

in the order they were created, as long as they are available to be executed.

If no tasks are available to be run, the Scheduler will run the idle task which

puts the CPU to sleep while waiting for an interrupt.

\subsection{Interrupt Controller}

\begin{figure}\begin{center}

\begin{bytefield}[endianness=big]{32}

\bitheader{0-31} \\

        \bitbox{1}{E}

        \bitbox{15}{Enabled}

        \bitbox{1}{A}

        \bitbox{15}{ACTIVE}

        \\

\end{bytefield}

\caption{Programmable Interrupt Control (PIC) Register}\label{fig:picreg}

\end{center}\end{figure}

This controller supports up to fifteen interrupts, however only twelve are

defined within the SoC.  If any interrupt line is active, the PIC controller

will have that bit set among it's active set.  Once set, the bit and hence the

interrupt can only be cleared by writing to the controller.  Interrupts can

also be enabled as well.  The enabled bit mask controls which interrupt lines

are permitted to interrupt the CPU. Hence, just because an interrupt is active

doesn't mean it will interrupt the CPU--the enabled line must be set as well.

Finally, then {\tt A} or {\tt ANY} bit will be high if any interrupts are both

enabled and active, whereas the {\tt E} or global interrupt enable bit can be

set to allow the PIC to interrupt the CPU or cleared to disable all interrupts.

To keep operations on this register atomic, most of the bits of this register

have special meanings upon write.  The one exception to this is the global

interrupt enable bit.  On any write, interrupts will be globally enabled or

disabled based upon the value of this bit.  Further, the {\tt ANY} bit is a

read only bit, so writes to it have no effect.

Enabling specific interrupts, via writes to the enable lines, are different.

To enable a specific interrupt, enable all interrupts and

set the wire high associated with the specific interrupt you wish to enable.

Hence writing a {\tt 0x80010000} will enable interrupt line zero, while also

enabling all previously enabled interrupts.

To disable a specific interrupt, disable all interrupts and write a one to the

enable line of the interrupt you wish to disable.  In this fashion, writing a

{\tt 0x00010000} will disable all interrupts and leave interrupt line zero

disabled when the interrupts are re--enabled later, whereas {\tt 0x07fff0000}

will disable all specific interrupts.

Interrupts are acknowledged in a fashion similar to enabling interrupts.  By

writing a `1' to the active bit mask, the interrupt will be acknowledged and

reset, whereas writing a `0' leaves the interrupt untouched.  In this fashion,

as individual interrupts are handled, a `1' may be written to this bottom mask

to clear the interrupt.  Be aware, however, that any interrupt acknowledgement

may also globally enable or disable interrupts.

\subsection{Last Bus Error Address}

Another use for this is upon any reboot, it is possible to read the address

of the last bus error and perhaps learn something of what caused the CPU to

restart.

\subsection{ZipTimer}

The S6 Soc contains two ZipTimers, available for the CPU to use.  These are

countdown timers.  Writing any non--zero value to them will cause them to

immediately start counting down from that value towards zero, and to interrupt

the CPU when they reach zero.  Writing a new value while the timer is running

will cause that new value to automatically load into the CPU and start counting

from there.  Writing a zero to the timer disables the timer, and causes it to

stop.

ZipTimer A can be set to auto reload.  When set, the timer will automatically

load it's last set value upon reaching zero and interrupting the CPU.  This

effectively turns it into an interrupt timer if desired.  To set this feature,

write to the timer the number of clock ticks before an interrupt, but also set

the high order bit.  In this fashion, writing a {\tt 0x80013880} will interrupt

the CPU every millisecond, starting one millisecond after the write takes place

(assuming an 80~MHz system clock).

ZipTimer B has been wired for a different purpose.  ZipTimer B does not support

auto reload, nor will it interrupt the CPU.  Instead, ZipTimer B has been wired

as a watchdog timer.  When this timer reaches zero, the CPU will be rebooted.

One way to use this timer would be in conjunction with the ZipTimer A, and to

write a number to it upon any entry to the interrupt service routine.  If given

enough time, this would cause the CPU to reboot if for any reason it locked up.

\bitheader{0-31} \\

        \bitbox{10}{Unused}

        \bitbox{1}{S}

        \bitbox{1}{G}

        \bitbox{3}{}

        \bitbox{1}{E}

        \bitbox{16}{Sample}

        \\

\end{bytefield}

\caption{PWM Audio Controller Bitfields}\label{fig:pwmreg}

\end{center}\end{figure}

This controller has been designed for easy writing.  To send a sample to the

PWM audio controller, simply write the sample to the controller and clear the

PWM audio interrupt.  When the audio interrupts the CPU again, it is ready

for the next sample.

The audio sample rate has been fixed at 8~kHz.  While changing this rate is

easy to do within {\tt busmaster.v}, the rate itself takes some work to keep

up with, so I wouldn't recommend going much (any) faster.

The audio controller supports two additional functionalities, however.  The

first is that the {\tt E} bit will be set upon any read when or if the audio

controller is ready for another sample.  Equivalently, the audio interrupt

will be asserted.

The second functionality has to do with the two auxiliary control bits present

in the PModAMP2 audio device.  These are the gain and shutdown bits.  To set

these bits, write a sample to the controller while also setting the {\tt E}

bit.  When the {\tt E} bit is set upon any write, the shutdown and gain bits

will also be set.  (Be aware, the shutdown bit is negative logic.)  Hence, one

may start this interface by writing a {\tt 0x0310000} to the device, and later

shut it back off by writing a {\tt 0x010000}.

As examples, writing a {\tt 0x0ff} to the {\tt SPIO} register will turn all

LED's on, {\tt 0x0f0} will turn all LED's off, and {\tt 0x011} and {\tt 0x010}

will turn LED0 on and then off again respectively.

means that, once generated, the interrupt must be disabled until the key or

button is released.

\subsection{General Purpose I/O}

will be generated.  There are no do not care lines.

output of either of these

\subsection{UART Data Register}

reading from the port has a real--time requirement associated with it--the

data buffer must be emptied before the next value is read.

be cleared while data is waiting.)  Writing to the UART port is done in a

similar fashion.  First, wait until the UART transmit interrupt is asserted,

second write to the UART port, and then third clear the interrupt.  As with

the read interrupt, clearing the interrupt prior to writing to the port will

have no effect.

\section{Debugging Scope}

The debugging scope consists of two registers, a control register and a data

register.  It needs to be internally wired to 32--wires, internal to the S6

SoC, that will be of interest later.  For further details on how to configure

and use this scope, please see the {\tt WBSCOPE} project on OpenCores.

\section{Internal Configuration Access Port}

The Internal Configuration Access Port (ICAP) provides access to the internal

configuration details of the FPGA.  This access was designed so as to provide

the CPU with the capability to command a different FPGA load.  In particular,

the code in Fig.~\ref{fig:reload} should reconfigure the FPGA from any given

Quad SPI {\tt address}.\footnote{According to Xilinx's technical support, this

will only work if the JTAG port is not busy.}

\begin{figure}\begin{center}\begin{tabbing}

{\tt warmboot(uint32 address) \{} \\

\hbox to 0.25in{}\={\tt uint32\_t *icape6 = (volatile uint32\_t *)0x{\em <ICAPE port address>};}\\

 \>{\tt icape6[13] = (address<<2)\&0x0ffff;}\\

 \>{\tt icape6[14] = ((address>>14)\&0x0ff)|((0x03)<<8);}\\

 \>{\tt icape6[4] = 14;}\\

 \>{\em // The CMod~S6 is now reconfiguring itself from the new address.}\\

 \>{\em // If all goes well, this routine will never return.}\\

{\tt \}}

\end{tabbing}

\caption{Spartan--6 ICAPE Usage}\label{fig:reload}

\end{center}\end{figure}

One subtle problem with this port is that it will not work if the CMod is

plugged in to the USB JTAG port.  It will only work if the CMod has been

provided with an independent power supply, leaving the USB JTAG unplugged.

For further details, please see either the {\tt WBICAPETWO} project on OpenCores

as well as Xilinx's ``Spartan-6 FPGA Configuration User Guide''.

\section{Real--Time Clock}

The Real Time Clock will be included if there is enough area to support it.

The four registers correspond to a clock, a timer, a stopwatch, and an alarm.

If space is tight, the timer and stopwatch, or indeed the entire clock, may be

removed from the design.  For further details regarding how to set and use this

clock, please see the {\tt RTCCLOCK} project on OpenCores.

There is currently not enough area on the chip to support the Real--Time Clock

together with all of the other peripherals listed here.  You can adjust whether

the clock is included or not by adjusting the {\tt `define} lines at the top

of {\tt busmaster.v}.  For example, it may be possible to get the RTC back by

disabling the ICAPE2 interface.

\section{On-Chip Block RAM}

The block RAM is the fastest memory available to the processor.  It is also

the {\em only} writeable memory available to the processor.  Hence all

non-constant program data {\em must} be placed into block RAM.  The ZipCPU

can also run instructions from the block RAM if extra speed is desired.  When

runnning from block RAM, the ZipCPU will nominally take 8~clocks per

instruction, for an effective rate of 8~MIPS.  Loads or stores to block RAM

will take one instruction longer.

\section{Flash Memory}

The flash memory has been arbitrarily sectioned into three sections, one for

a primary configuration, a second section for an alternate configuration file,

and the third section for any program and data.  These regions are shown in

Tbl.~\ref{tbl:flash-addresses}.

\begin{table}[htbp]

\begin{center}\begin{tabular}{|p{0.75in}|p{0.75in}|p{0.5in}|p{3.0in}|}\hline

\rowcolor[gray]{0.85} Start & End & & Purpose \\\hline\hline

\scalebox{0.9}{\tt 0x400000} & \scalebox{0.9}{\tt 0x43ffff} & R & Primary configuration space\\\hline

\scalebox{0.9}{\tt 0x440000} & \scalebox{0.9}{\tt 0x47ffff} & R & Alternate configuration space\\\hline

\scalebox{0.9}{\tt 0x480000} & \scalebox{0.9}{\tt 0x7fffff} & R & ZipCPU program memory\\\hline

\end{tabular}

\caption{Flash Address Regions}\label{tbl:flash-addresses}

\end{center}\end{table}

The host program {\tt zipload} can be used to load a ZipCPU program and

configuration files into this address space.  To use it, first load the

alternate configuration into the FPGA.  Then pass it, as arguments, the

primary, and alternate if desired, configuration files followed by the ZipCPU

program file.  Then, when the primary configuration is loaded again, perhaps

upon power up, the ZipCPU will automatically start running from it's

{\tt RESET\_ADDRESS}, {\tt 0x480000}.

When running from Flash memory, the ZipCPU will nominally take 52~clocks per

instruction, for an effective speed of about 1.5~MIPS.