[U-Boot-Users] Zilog Z85230 serial communication controller
Hi again!
Currently I try to integrate a Zilog Z85230 serial communication controller on a 603e-based board to work with U-Boot. For testing reasons I wrote a little inline assembler program looking like that:
#include <common.h>
#ifdef CFG_Z85230
#if CONFIG_CONS_INDEX == 1 #else #error no valid console defined #endif
int serial_init (void) {
/* First we load the _physical_ address of the chip's cntrol register to R3 */ __asm__ __volatile__ ("addis 3, 0, 0x0000"); __asm__ __volatile__ ("addis 3,0,0xF500"); __asm__ __volatile__ ("ori 3,3,0x00D0"); /* Load the address of the chip's data rgister to R2 */ __asm__ __volatile__ ("addis 2, 0, 0x0000"); __asm__ __volatile__ ("addis 2,0,0xF500"); __asm__ __volatile__ ("ori 2,2,0x00D8");
/* Here some configuration for the serial controller is done, use R4 */ __asm__ __volatile__ ("li 4, 0x04"); __asm__ __volatile__ ("stb 4, 0(3)"); __asm__ __volatile__ ("li 4, 0x44"); __asm__ __volatile__ ("stb 4, 0(3)");
. . . .
/* Write one character ('I') */ __asm__ __volatile__ ("addis 4, 0, 0x0000"); __asm__ __volatile__ ("li 4, 0x49"); __asm__ __volatile__ ("stb 4, 0(2)");
return(0); } . . .
When I compile this, load it into RAM and execute it, everything is fine, MiniCom shows an 'I'. Now I put this litte function into the file serial.c from U-Boot and compile it. Then I load U-Boot to Flash and run it. I'm tested before, wheather the function is always called and it is.
When I run it, I get an error, caused by the MMU, because it cannot translate the demanded address (it's a physical one). Now I tried to disable the data-address-translation on the beginning of the function and re-enable it at the end, always surrounded by an 'isync'-command. It this case, the program runs, but without any output via the serial port.
Is it legal to disable the MMU in this context and how should I configure U-Boot for my memory map?
I have the following memory map on my board:
0x00000000 - 0x007FEFFF SRAM
0xF5000000 - Serial Interface #1 Channel B Control 0xF5000008 - Serial Interface #1 Channel B Data . . . 0xF50000D0 - Serial Interface #3 Channel A Control 0xF50000D8 - Serial Interface #3 Channel A Data
0xFF800000 - 0xFFFFFFFF Flash
Thanks for ideas or advices why this doesn't work?
Sebastian Häpe
Häpe, Sebastian, HRD/AB wrote:
Hi again!
Currently I try to integrate a Zilog Z85230 serial communication controller on a 603e-based board to work with U-Boot.
[snip]
When I compile this, load it into RAM and execute it, everything is fine, MiniCom shows an 'I'.
I assume you are putting a stand-alone snippet of code in RAM -- in this case, your MMU and caches are disabled.
Now I put this litte function into the file serial.c from U-Boot and compile it. Then I load U-Boot to Flash and run it. I'm tested before, wheather the function is always called and it is.
When I run it, I get an error, caused by the MMU, because it cannot translate the demanded address (it's a physical one). Now I tried to disable the data-address-translation on the beginning of the function and re-enable it at the end, always surrounded by an 'isync'-command. It this case, the program runs, but without any output via the serial port.
Is it legal to disable the MMU in this context and how should I configure U-Boot for my memory map?
I have the following memory map on my board:
0x00000000 - 0x007FEFFF SRAM
0xF5000000 - Serial Interface #1 Channel B Control 0xF5000008 - Serial Interface #1 Channel B Data . . . 0xF50000D0 - Serial Interface #3 Channel A Control 0xF50000D8 - Serial Interface #3 Channel A Data
0xFF800000 - 0xFFFFFFFF Flash
Thanks for ideas or advices why this doesn't work?
Sebastian Häpe
It doesn't run from u-boot because you have not mapped the physical memory space using a BAT (or other MMU method). You need to map the memory area _and mark it uncached_.
By disabling the MMU around the call, you got around the memory protection issue, but now you have a caching issue: u-boot will be running with data caching enabled (and instruction caching, but that shouldn't be a problem). When you do your write to the control registers you are only writing to cache, not to the registers themselves. There is no reason for the CPU to write your register-bound data to the registers, so you never see anything coming out.
The only way to make this work is to understand and use the MMU and cache control or disable them altogether (hint: disabling is not a good choice other than proof-of-concept testing :-).
* Do some User Manual reading on your processor, concentrating on the sections covering MMU (DBATs) and cache.
* Figure out which DBAT(s) are available for use and use one of them to map your peripheral space.
gvb
In message 2D9AB865A49FD511914700105A0EF67901C6F198@morse.smart.lfk.eads.net you wrote:
Currently I try to integrate a Zilog Z85230 serial communication controller on a 603e-based board to work with U-Boot.
Jerry van Baren already explained you the core problem (memory mapping) of your approach.
For testing reasons I wrote a little inline assembler program looking like that:
My question is: why do you write this in assembler? There is no reason to do this. Please write your code in C if you hope to see it added to U-Boot one day.
Viele Grüße,
Wolfgang Denk
participants (3)
-
"Häpe, Sebastian, HRD/AB" -
Jerry Van Baren -
Wolfgang Denk