AVR Freaks

Thanks for your quick reply, glitch.

>> 1) compile the 3 pieces of code separately...

Unfortunately, I can't compile all of them separately, even though ultimately they are to be programmed separately. Also, they are not 3 separate applications as is commonly the case with bootloaders.

It's somewhat complex, but there are actually 3 different loaders, 2 of which share some code. One is a typical bootloader, the one that gives no problems. Another is a later-added DPRAM-bound loader, and there was yet another request for a bootloader and DPRAM-loader programmer (yes, it sounds just as crazy to me as it does to you, but it was not my decision).

There is some programming code in the app area, with the actual page writer in the boot area (necessarily), which must be completely isolated from the plain bootloader in the boot area, so it can't be overwritten when the bootloader is reprogrammed remotely. There is some RAM shared between 2 of the programmers, which is what makes compiling them separately difficult.

>> 2) correct your segmentation, since you have apparantly forced 3 different address spaces instead of just one. This will require editing & correcting your XCL file.

Here are the relevant entries in one of the XCLs I've tried. I can't get the compiler to assign absolute addresses to the C routines, so I fudged this. If you have any ideas...

-Z(CODE)NEAR_F,HUGE_F,SWITCH,INITTAB,DIFUNCT,CODE=0-FLASH_LAST
-Z(FARCODE)FAR_F=BOOT_BASE-FLASH_LAST
-Z(CODE)BOOT_CODE_AREA=(BOOT_BASE+IVT_SIZE)-FLASH_LAST
-Z(CODE)LAST_BOOT_PAGE=(FLASH_SIZE-100)-FLASH_LAST
-Z(XDATA)EEPROM_N=0-FFF

I don't have enough time to revert to pure assembler, and I must ensure some code goes only into the LAST_BOOT_PAGE.

Thanks for your help

Hi pepsi,

That's one I didn't think of!

Unfortunately, the client needs to be able to recompile the code off-site, and then remotely upload it to the device with the AVR. In fact, that's what caused all the problems: the original serial bootloader (which they felt was insufficient), then the DPRAM loader for remote programming, and then the killer, another loader to reprogram the other 2 loaders! (in case of future modifications or bugs)

However, I'll use your idea to test the third loader, since I can move the chip into the test board, burn the binary with the JTAG, then pull the hex out.

Thanks