Reply by Keith E. Larson ●January 4, 20022002-01-04
Hi Rick

I put togther this example last year showing how to bit bang out the correct
serial port patterns using the PC printer port data lines as the drivers. I
did not work out the receive side as I had no intention of porting the DSK
drivers and debugger, but I suppose it could also be done with a little effort.

Serial Booting the VC33 and C31
from the PC printer port

The following is taken from a message and has not been fully massaged into
an application note. It took me about 5 hours to slap together the circuit
and the code, so it may be a little rough on the edges. Never the less it
works. Note that both the host side and DSP side code are given and that I
used the DSK tools to generate the HEX file (much much easier to use!). Here
are some other questions and answers that may come up.

The receiving serial port (bootside) should be inactive when reset
occurs. This simply means the FSR is inactive as the clock and data are dont
care unless a frame start is detected.

The serial port boot mode skips over the first two words of a normal boot
from EPROM file. Dont ask why, its just that way and has been for a very
long time. Incidentaly, note how stupid it was to even include the bus
control in the header. This is a memory mapped register and can therefor be
loaded as a data block. So the bootloader code could have been even simpler!

The send side transmitter should be in the same state as the receiver
except for the clock edge. Send and receive should use opposite edges
unless you need to run at very fast rates in a bad environment in which case
you will need to have some control over the skew. The configuration is
given in the bootloader source code.

Minimum pulse widths are 2*H when your external sources are synchronous
to the receiving CPU's master clock. This is a kind of Nyquist limit that
is imposed by the fact that each pin input is first passed through a set of
synchronizing latches. If you are not synchronous, you need to be below
the Nyquist rate. The data sheet shows 2.6*H, but the hard cold truth is
that this is actualy 2.00001*H ... if the H clock rate is very slow. At
speed the setup and hold times become significant enough to warrant the 2.6.

PS: The TMX (PG version 1.0 and 1.1) VC33's have a slight problem with
delays in the serial port that does limit the speed when used at min
volatage, max speed and high temperature. This wont be a problem for a
printer port link and is fixed in the production version (PG 1.2) currently
shipping.

There is no maximum width when configured in this mode since the clock
is external. All you need to ensure is that the data bit is clean and
stable at the time the sample is taken. In my code I ran the delay all the
way to 50ms. This was teadiously slow even for this very short program.
Please note however that the 'delay()' function has given me some problems
in WIN98. I dont know exactly whats going on, but I suspect that Windows is
trying to mis-manage the timers. So... I simply created my own delay the
old fashioned brute force kind of way.

Serial port receive configuration
(from bootloader code comments)
Recieve control: 0x111 (all pins are serial port pins)
Global control : 0xA300000

RRESET = 1 Reset receive port
RINT = 1 Enable RINT (will be polled in boot s/w)
RTINT = 0 Recieve timer off
RLEN = 3 32 bit
FSRP = 0 active high
DRP = 0 active high
CLKRP = 0 active high
RFSM = 0 Standard mode (non continuous, use FSX).
RVAREN = 0 Fixed data rate signaling (Falling edge starts xmit)
RCLKSRC = 0 External
RSRFULL = 0 Polling bit
RRDY = 0 Polling bit Expected waveform
______ _______________
FSR \_______________________________/////// \___________
_______ _______ _______ _______ _________ __
DR XXXXXX__MSB__X_______X_______X__LSB__XXXXXXXXXXXXXXXXX___MSB___X__
_ ___ ___ ___ ___ ___ ___ ___ ___
CLKR \___/ \___/ \___/ \___/ \___/ \___/ \___/ \___/ Making the connection

The circuit construction I used is quite simple. I simply took a male DB25
connector and soldered in a 13x2 array of extra long wire wrap posts. I had
to bend them out a little bit and clip off the 26th contact. I then clipped
out the posts for D0-D5 leaving a little stub to solder to and replaced them
with 330 ohm resistors (100 should be even better but I did not have that
value in my desk drawer). I then used jumper wires with female wire-wrap
post connectors on each end to make the connections to my test board.

Note: The TMS320VC33 uses Schmidt Hysterisis inputs for all inputs except
for the data bus. This helps considerably in quality signal reception.

Host Side Code

/*--------------------------------
Booting a C31/C33 from a printer port
Keith Larson
(c) Texas Instruments Inc, 2001

VC33 connections to printer port are as follows. Add a series
resistor 100-470 ohms in series to prevent accidental buffer
contention since each of the serial port lines can be configured
as an output.

D0 - CLKR
D1 - DR
D2 - FSR
D3 - RESET
D4 - INT3

Note: If you want to create a logic waveform you can modify the
outport() function to also print in binary to a file or
screen dump.
----------------------------------*/
#include <conio.h>
#include <stdio.h>
#include <stdlib.h>
#include "dsk.h"

#define CLKR (clkr^ 1) // 1 CLKRP
#define DR (dr ^ 0) // 2 DRP
#define FSR (fsr ^ 0) // 4 FSRP

ulong boot_data[256];
uint port = 0x378; // LPT1 data address, 0x278 for LPT2
int RESET = 1;
int INT3 = 1;
int dr, clkr, fsr;
int dat;
char buf[82];
//
// If this program is to run in a Windows DOS box
// use this delay instead of the normal delay function
// There is some kind of bug in WIN98. Apparently in
// how the timers are interrupted.
#define DELAY 0 // Try 50+ for very slow!
void mydelay(int ms);
void ReadHexFile(ulong *d, char *Name);
//
void serial_out(ulong val)
{
int s;
ulong v;
printf("Sending: %08lx ",val);
clkr = 0<<0;
dr = 0<<1;
fsr = 1<<2; // FSR is active for more than 1 CLKR cycle
for(s=0;s<2;s++)
{
dat = FSR|DR|CLKR; outport(port,0x18|dat); clkr^=1; // Clock twice with
FSX=1
}
mydelay(DELAY);
for(s1;s>-1;s--)
{
fsr = 0<<2; // FSR goes back to inactive on 1st bit
v=(val>>s)&1;
dr = (int)v;
printf("%d",dr);
dr = dr<<1;
dat = FSR|DR|CLKR;outport(port,0x18|dat); clkr^=1;
dat = FSR|DR|CLKR;outport(port,0x18|dat); clkr^=1;
mydelay(DELAY);
}
printf("\n");
}

void main(void)
{
long L,A;
ulong *sd;
clrscr();
ReadHexFile(boot_data, "xf_togl.hex");
clkr = 0<<0; // clkr is inactive
dr = 0<<1;
fsr = 0<<2; // fsr is inactive
dat = FSR|DR|CLKR; dat|=0x00; outport(port,dat); printf("I=0 R=0
%04x\n",dat);
dat = FSR|DR|CLKR; dat|=0x08; outport(port,dat); printf("I=0 R=1
%04x\n",dat);
dat = FSR|DR|CLKR; dat|=0x18; outport(port,dat); printf("I=1 R=1
%04x\n",dat);
printf("Boot start\n");
mydelay(DELAY);
//
// Now running bootloader in serial mode
//
sd = boot_data+2; // skip memwidth and bus_control
for(;;)
{
L = *sd++; // send block length
serial_out(L); // zero terminates boot
if(L==0) break; // if zero, done
A = *sd++; // send block address
serial_out(A);
for(;L>0;L--) // send block
{
serial_out(*sd++);
}
}
printf("Boot complete\n\n");
}
//===================================================
// If this program is to run in a Windows DOS box
// use this delay instead of the normal delay function
// There is some kind of bug in WIN98
void mydelay(int ms)
{
int t;
for(;ms>0;ms--)
{
for(t=0;t<500;t++)
inport(port+1);
}
}
void ReadHexFile(ulong *d, char *Name)
{
int cnt;
char *c,*eptr;
FILE *file;
printf("preparing to read: %s\n",Name);
file = fopen(Name,"rb");
if(file==NULL) { printf("cant open %s",Name); exit(1);}
c = fgets(buf,80,file);
if(c==NULL) { printf("invalid hex file\n"); exit(1);}
cnt = 0;
for(;;)
{
c = fgets(buf,80,file);
if(c==NULL) break;
*d++ = strtol(buf,&eptr,0);
cnt++;
}
printf("%d lines read\n\n",cnt);
}

Assembly Code I used for Testing the Bootloader

;***************************************************************
; XF_TOGL.ASM
; Keith Larson
; (c) Texas Instruments Inc, 2001
;
; Toggles the XF0 and XF1 pins (and the IACK pin) at the loop
; rate. Add nops or some other kind of delay to change the
; pulse rates and widths.
;
; NOTE: Avoid 0x809800 and 0x809801 as these locations
; are used by the bootloader as stack space
;***************************************************************
.start "CODE",0x809802
.sect "CODE"
.entry MAIN
MAIN LOPOWER
ldp @0x808064
ldi 0x1018,R0
sti R0,@0x808064
LOOP1 ldi 1000,RC
RPTB LOOP2
ldi 0x22,IOF
ldi 0x66,IOF
LOOP2 iack @0x80A000
b LOOP1

*********** Resulting DSK generated Hex file *********
FILE2HEX
0x00000008 Bus Width <- Skip this in SP boot
0x000010f8 Bus Control ws=7 <- Skip this in SP boot
0x00000001 Dummy Section size
0x00809802 Dummy Start Address
0x00000000 Dummy Data
0x0000000a Section Size
0x00809802 Start Address 'CODE'
0x10800001
0x50700080
0x08601018
0x15208064
0x087b03e8
0x6480980a
0x08780022
0x08780066
0x1b20a000
0x6a00fffa
0x00000000 Terminate Load Anyone had experience with booting a C31 DSP via the DSP serial port?
Specifically I am looking for PC code to control parallel port IO lines
under Windows 95 to let me boot a DSP in order to program the flash on the
board.

We had been doing this via the emulator, but the pod just quit. Rather than
continue doing it this way, I would rather put the effort into getting the
boards up via the DSP serial port. The emulator GEL scripts never worked
very well. They would sometimes run correctly and sometimes not.

I have code to run on the DSP to let me program the flash if I can get the
program into the ram on board.

Rick Collins
Arius - A Signal Processing Solutions Company
Specializing in DSP and FPGA design http://www.arius.com
4 King Ave 301-682-7772 Voice
Frederick, MD 21701-3110 301-682-7666 FAX