Re: [avr-gcc-list] VMA & LMA address -- confused ?

[Erik Christiansen]

On Thu, Dec 09, 2010 at 03:57:44PM +0530, Dinesh Guleria wrote:
> I am bit confused about VMA & LMA address for AVR.
> Virtual memory address & Load Memory Address.
> 1> What actually does it means ???????

The VMA is actually the runtime-address of the section. e.g. If code is
to be loaded into ROM, copied to RAM. and then executed, then VMA !=
LMA. That does not happen on the AVR, but non-zero initialised data does
experience this transfer.

> What i got from reading is that VMA is passed to linker & .ELF is created.

Please note that every section has both VMA and LMA. (Though they may be
the same.) All sections can be expected to be loaded to their LMA.

> While avr-objcopy read .elf & creats the .hex  in this hex file .crt code
> will place the data variable to LMA. & .eep file in this file epprom
> variables are placed to LMA.

The LMA to VMA copying performed by .crt code is only non-zero initialised
variables. Note: .crt code copies _from_ LMA.

If you're using unmodified AVR toolchain linker scripts, then it is
likely that everything will work fine, and you can happily observe what
VMA and LMA are set to.

> Because of harvard architectue to avoid confusion related to memory
> address ( because EEPROM, RAM & Flash all will start from 0x0000 i.e
> ZERO address) GCC does following :--- Add 0x800000 the to real SRAM
> --- if you want the .data section to start at 0x1100, pass 0x801100 at
> the address Add 0x810000 the to real EEPROM 2> What does it add for
> the ROM address, i am bit confused --- ???????? 3> Also why this
> typical number series are chosen 0x800000 .... ???????

These offsets exist only to allow gcc to handle the harvard architectue
when it was never designed to do so. i.e. It is a workaround. Once you
realise that the offsets are there to fool gcc, it is easier not to
fooled as well.

The numbers only need to be large enough to avoid section overlap in all
anticipated future AVR devices. ;-)

> This operation produces VMA. This is only my understanding &
> assumption.

Nah. ;-)
Here you can see that both LMA and VMA are subject to the offsets:

$ avr-objdump -h test.elf

test.elf:     file format elf32-avr

Sections:
Idx Name          Size      VMA       LMA       File off  Algn
  0 .text         0000003a  00000000  00000000  00000094  2**0
                  CONTENTS, ALLOC, LOAD, READONLY, CODE
  1 .data         00000000  00800100  0000003a  000000ce  2**0
                  CONTENTS, ALLOC, LOAD, DATA
  2 .bss          00000000  00800100  0000003a  000000ce  2**0
                  ALLOC
  3 .noinit       00000000  00800100  00800100  000000ce  2**0
                  CONTENTS
  4 .eeprom       00000000  00810000  00810000  000000ce  2**0
                  CONTENTS

You can also see that "What does it add for the ROM address"? is zero.
(For .data, it may appear that only VMA is affected by offset, but that
is only because the offset for ROM is zero.)

There are enough hints in this reply to guide further repeated
reading of whatever documentation you can find. (I've not come across a
lot, but the gnu ls "info" page can be mined for understanding, with
repeated reading, interspersed by experiments.)

The best way to learn, I've found, is to write linker scripts, and tweak
them until they do what you want. (Admittedly, other targets than AVR
provide more scope for fiddling with VMA != LMA)

> 
> Please correct me for the above three question.

These few answers will hopefully help toward clearing the confusion.
At least they'll shatter a few false assumptions, which will allow
clearer understanding of what you read.

It is all confusing at first.

Erik