The following code fragment has been running for three years on an Atmel 89c55 with no problems. Now that I have started a project using the Cypress EZ-USB FX chip I would like to move the fifo buffer from the default generic pointer in data space. #define FIFO0SIZE 10 BYTE *fifo0[FIFO0SIZE]; to xdata and increase the size using #define FIFO0SIZE 100 xdata BYTE *fifo0[FIFO0SIZE]; It compiles but continualy sends garbage out the serial port. I have tried pointer casting and changing to the large memory modle and a few other things, all with negative results. Can anyone shed some light on this problem. #define FIFO0FULL FIFO0SIZE-2 BYTE Que(new_ptr) BYTE *new_ptr; { BYTE idata Pass,i; Pass=0xff; for(i=0;i<FIFO0SIZE-1;i++){ // add new message to end of buffer if(!fifo0[i]){ // current position is empty go ahead and use it if(i>FIFO0FULL) SBuff0_Full=1; Pass=0; fifo0[i]=(BYTE xdata *) new_ptr; fifo0[i+1]=0; if(!out0_busy) TI=1; // kick start the uart break; } } return Pass; } void Serial_Port0(void) interrupt COM0_VECT { BYTE idata i; static BYTE data x=0; if(RI) { RI=0; // extract recieved byte Inkey(SBUF0); } if(TI) { TI=0; if(fifo0[0][x]) { SBUF0=(fifo0[0][x]); x++; out0_busy=1; } else out0_busy=0; // no character to send if(!fifo0[0][x]) { // end of string last character is being sent if(out0_busy) { // last character is being sent for(i=0;i<FIFO0SIZE-1;i++) { // rotate buffer if(fifo0[i]) { // if = 1 = buffer not empty fifo0[i]=fifo0[i+1]; x=0; } // if fifo0 buffer not empty else { if(i<2) SBuff0_Full=0; break; // end of buffer } } } } } }
I could be misinterpreting what your code is trying to do, but... fifo0 is defined as an xdata array of pointers to BYTE's that reside in the data memory space. I'm making the assumption that the message buffers reside in xdata also. If this is the case, your array elements should be xdata pointers. I think you'd need to change fifo0's defining declaration to:
xdata BYTE xdata *fifo0[FIFO0SIZE];
BYTE Que(new_ptr) BYTE xdata *new_ptr; { /* ... */
Oops, nevermind. Kindly ignore my previous post. I had read your assembly listing excerpt too hastily and jumped to an incorrect conclusion. Sorry. --Dan Henry
Thank you for your help. This is an ASM code fragment that has been running for three years on an Atmel 89c55 with no problems. Now that I have started a project using the Cypress EZ-USB FX chip I would like to move the fifo buffer from the default generic pointer in data space to a generic pointer stored in xdata and increase the buffer size. This was the only change I thought was necessary. from
#define FIFO0SIZE 10 BYTE *fifo0[FIFO0SIZE];
#define FIFO0SIZE 100 xdata BYTE *fifo0[FIFO0SIZE];
; SOURCE LINE # 451 0031 900000 R MOV DPTR,#fifo0 0034 E0 MOVX A,@DPTR 0035 FB MOV R3,A 0036 A3 INC DPTR 0037 E0 MOVX A,@DPTR 0038 FA MOV R2,A 0039 A3 INC DPTR 003A E0 MOVX A,@DPTR 003B F9 MOV R1,A 003C 850082 R MOV DPL,x 003F 758300 MOV DPH,#00H 0042 120000 E LCALL ?C?CLDOPTR 0045 FF MOV R7,A 0046 6008 JZ ?C0076 ; SOURCE LINE # 452 0048 F599 MOV SBUF0,A
; SOURCE LINE # 451 0031 AB00 R MOV R3,fifo0 0033 AA00 R MOV R2,fifo0+01H 0035 A900 R MOV R1,fifo0+02H 0037 850082 R MOV DPL,x 003A 758300 MOV DPH,#00H 003D 120000 E LCALL ?C?CLDOPTR 0040 FF MOV R7,A 0041 6008 JZ ?C0076 ; SOURCE LINE # 452 0043 F599 MOV SBUF0,A
I agree that the assembly code fragments look like they should work. So that kinda makes me want to look elsewhere. BTW, did you ever have your "original" version (before moving to xdata) working on the Cypress part? That Cypress part is a speedy little bugger. It looks like CLK24 or CLK48 feeds the timers you'd be using for baud rate generation. Can we eliminate UART and BRG initialization differences factoring into the equation? I guess back to the first question -- had code running on the Cypress part ever output non-garbage via the UART (even the simplest single character non-interrupt driven output)? --Dan Henry
Yes, the NON xdata (original) version works fine on the EZ-USB FX. The reason I MUST change to xdata is the USB code uses more of the internal RAM than my original code can spare. I'm sure the problem is very simple. Once I figure it out I will be embarrassed that I was stumped for so long. My problem is I have been using the interrupt driven Queuing scheme for so long I may be making fasle assumptions. My eventual goal is to convert the pointer to pointer buffer to a rotary buffer of the actual characters to be sent out the serial port. With the old system I was limited to internal memory. Now with the Cypress EZ-USB FX I can make this move. But my Que will still be receiving messages from DATA, IDATA and CODE. If this is the problem I am running into i.e. Generic pointers from different MSpaces. I may be fighting the same problem with the added confustion of all new code. Do you know of a good xdata rotary buffer example I could look at. Thank you very much for your input. Bob Dillon
To satisfy my curiosity, I tried simulating your code in its "xdata" version. I ran into a couple of problems, one of which could have been due to the simulation environment. Quoting your original code excerpts for reference:
BYTE Que(new_ptr) BYTE *new_ptr; { BYTE idata Pass,i; Pass=0xff; for(i=0;i<FIFO0SIZE-1;i++){ // add new message to end of buffer if(!fifo0[i]){ // current position is empty go ahead and use it if(i>FIFO0FULL) SBuff0_Full=1; Pass=0; fifo0[i]= (BYTE xdata *) new_ptr; // <-- ?1? fifo0[i+1]=0; if(!out0_busy) TI=1; // kick start the uart break; } } return Pass; } void Serial_Port0(void) interrupt COM0_VECT { BYTE idata i; static BYTE data x=0; if(RI) { RI=0; // extract recieved byte Inkey(SBUF0); } if(TI) { TI=0; if(fifo0[0][x]) { SBUF0=(fifo0[0][x]); x++; out0_busy=1; } else out0_busy=0; // no character to send if(!fifo0[0][x]) { // end of string last character is being sent if(out0_busy) { // last character is being sent for(i=0;i<FIFO0SIZE-1;i++) { // rotate buffer if(fifo0[i]) { // if = 1 = buffer not empty fifo0[i]=fifo0[i+1]; x=0; // <-- ?2? } // if fifo0 buffer not empty else { if(i<2) SBuff0_Full=0; break; // end of buffer } } } } } }
BYTE *byte_ptr; /* Advances through message */ BYTE **tail_ptr = fifo0; /* Where to enqueue next msg ptr */ BYTE **head_ptr = fifo0; /* Where to dequeue next msg ptr */ BYTE queued_ctr; /* Count of msg's queued in FIFO */ BYTE Que(new_ptr) BYTE *new_ptr; { BYTE Pass=0xff; /* Don't enqueue NUL messages. */ if (*new_ptr != 0) { /* If the FIFO queue has room, enqueue new_ptr at * the tail. Disabe serial interrupts while * accessing FIFO queue controls. */ ES0 = 0; if (queued_ctr != FIFO0SIZE) { queued_ctr++; *tail_ptr = new_ptr; /* Advance the tail, wrapping to the start of * the FIFO array if necessary. */ if (++tail_ptr == &fifo0[FIFO0SIZE]) tail_ptr = fifo0; Pass = 0; } ES0 = 1; /* out0_busy is FALSE only when the transmit ISR * has nothing to transmit (e.g., after either * init or transmitting the last message's last * non-NUL byte). */ if (!out0_busy) { out0_busy = 1; /* Will be busy after TI is set. */ byte_ptr = *head_ptr; TI = 1; } } return Pass; } void Serial_Port0(void) interrupt COM0_VECT { BYTE b; if(RI) { RI=0; // extract recieved byte Inkey(SBUF0); } if (TI) { TI = 0; b = *byte_ptr++; if (b) SBUF0 = b; else { /* At end of message. If more messages in the * FIFO, advance the head to the next one * (wrapping if necessary) and start * transmitting from there. Otherwise, * indicate that the transmit ISR has shut * down. */ if (--queued_ctr) { if (++head_ptr == &fifo0[FIFO0SIZE]) head_ptr = fifo0; byte_ptr = *head_ptr; SBUF0 = *byte_ptr++; } else out0_busy = 0; } } }
1. If someday some code should be added between
ES0 = 1;
if (!out0_busy)
if (!out0_busy) { out0_busy = 1; /* Will be busy after TI is set. */ byte_ptr = *head_ptr; TI = 1; } ES0 = 1;
if (--queued_ctr) { if (++head_ptr == &fifo0[FIFO0SIZE]) head_ptr = fifo0; byte_ptr = *head_ptr; SBUF0 = *byte_ptr++; } else out0_busy = 0;
if (++head_ptr == &fifo0[FIFO0SIZE]) head_ptr = fifo0; if (--queued_ctr) { byte_ptr = *head_ptr; SBUF0 = *byte_ptr++; } else out0_busy = 0;
Thank you all for your help, Andrew Neil, Dan Henry and Bob Mu. I do appreciate the time and effort you spent looking at my code and making suggestions. Dan, you are correct about ?1? the mspace casting and the loss of the Generic Pointers. It was an artifact from some of the code changes I was making while I was desperately trying to figure this out. As to ?2? the variable x is properly maintained in the hardware version. Note: This did inspire me to go back to my original system that this Serial routine was taken from and change the uC from an Atmel 89c55 to a Philips 89c51rd+ which has additional internal RAM accessed with a register bit and xdata/movx. My original code on the original hardware does work with both data and xdata buffers. Dan may be right, my problem may be related to the EZ-USB FX. Another Note. I do agree that a circular buffer is the traditional way to go. My problem was I was out of RAM. A circular buffer requires the Head and Tail pointer variable in addition to the buffer. If I used one more byte of RAM the code crashes due to Stack conflict. The original system communicated through the serial port at 300 baud to a system that did not use a UART. Instead it used a pin on a uP. Therefore not only was it a low baud rate but I had to put a delay between messages while it processed the previous message. The initial message is so long it takes about 3 minutes at 300 baud to complete the down load. All of these requirements led me to this style of serial communications. Bob you are correct about ES0 = 1. I had already made the change you first suggested and narrowed the last problem down to the if(--queued_ctr) . The problem was as long as you continue to stuff the buffer and never let it run out, the messages continue to flow. But as soon as you let queued_ctr run down to zero you never get the last message. Particularly troublesome is when you place a short message in the buffer and let a message finish before you try to send another short message you will get the old message out over and over until you stuff the buffer faster than it can be depleted. Here is the final version of the circular buffer. Note: There are a few global variables I need to maintain for other functions to use.
#define FIFOSIZE 100 xdata BYTE *fifo[FIFOSIZE]; BYTE *byte_ptr; /* Advances through message */ BYTE **tail_ptr = fifo; /* Where to enqueue next msg ptr */ BYTE **head_ptr = fifo; /* Where to dequeue next msg ptr */ BYTE queued_ctr=0; /* Count of msg's queued in FIFO */ BYTE Que(new_ptr) BYTE *new_ptr; { BYTE idata Pass; bit bdata EAstat, ESstat; EAstat=EA; EA=0; /* disable all int */ ESstat=ES; ES=0; /* turn off serial int */ Pass=0xff; if (*new_ptr != 0) { /* If the FIFO queue has room, enqueue new_ptr at * the tail. Disable serial interrupts while * accessing FIFO queue controls. */ if (queued_ctr++ < FIFOSIZE) { *tail_ptr = new_ptr; /* Advance the tail, wrapping to the start of * the FIFO array if necessary. */ if (++tail_ptr == &fifo[FIFOSIZE]) tail_ptr = fifo; if(queued_ctr>=FIFOSIZE) SBuff_Full=1; Pass = 0; } /* out0_busy is FALSE only when the transmit ISR * has nothing to transmit (e.g., after either * init or transmitting the last message's last * non-NUL byte). */ if (!out_busy) { out_busy = 1; /* Will be busy after TI is set. */ byte_ptr = *head_ptr; TI = 1; } } ES=ESstat; EA=EAstat; return Pass; } /* **************** UART I/O *********************** */ void Serial_Port0(void) interrupt 4 { BYTE b; if(RI) { RI=0; // extract received byte Inkey(SBUF); } if (TI) { TI = 0; b = *byte_ptr++; if((*byte_ptr==0)) { ES=0; Throttle(25); /* delay = 25 x DelayTick*/ /* 1/4 Sec = 25 x 10mSec */ TR0=1; /* Turn on timer 0 */ ET0=1; /* Enable Timer 0 interrupts */ } if (b) { SBUF = b; out_busy=1; } else { /* At end of message. If more messages in the * FIFO, advance the head to the next one * (wrapping if necessary) and start * transmitting from there. Otherwise, * indicate that the transmit ISR has shut * down. */ out_busy = 0; if (++head_ptr == &fifo[FIFOSIZE]) head_ptr = fifo; if (--queued_ctr) { byte_ptr = *head_ptr; SBUF = *byte_ptr++; out_busy = 1; } } } }