Hi
I am programming to an embedded ARM platform and my question is, is it possible to force constants to stay in the code segment (constant strings etc.) In C you can do this using "static". It looks as all string constants are placed in the data segment, and all variables are placed in the bss segment. Is that always the case?

Regards Carsten

Yes. What's the exact reason you need them in text segment?

In embedded work *all* should be possible. Every job is a special job and needs special solutions.

Da kommt noch einiges auf euch zu! ;-)

Martin

The Arm I use has 512KB flash and 32KB Ram, therefore I would like if I could define text constants that where placed in the flash segment (text). A way around this is:
1. To define string constants in an assembler file.
2. To tell the linker that the data segment are placed in flash (at the right location). Then change the start up code so it dos not initialize the data segment and change the system files so all predefined variables are changed to normal variables which are initialized. I have not tried (2), but I assume it is possible.

The very best would be if things was like this

Static
St: array[0..4] of char ='12345'; {Placed in text segment}

Var
St: array[0..4] of char ='12345'; {Placed in data segment}

Or something that worked the same way.

Regards
Carsten

Or to have any count of named segments which can be selected at var or const definition.

Var in text <- use the appropriate notation St: array[0..4] of char ='12345'; {Placed in text segment}

Var in bss
St: array[0..4] of char ='12345'; {Placed in data segment}

Const in myspecialseg
St: array[0..4] of char ='12345'; {Placed in special segment}

Martin

Hi

Would there be a chance to implement "Static" or something that did the same trick, in the compiler. Now I write my constants in an assembler file. That´s acceptable, but it is not a very elegant way to do it.

Regards Carsten

Would there be a chance to implement "Static" or something that did the same trick, in the compiler. Now I write my constants in an assembler file. That?s acceptable, but it is not a very elegant way to do it.

It's not the keyword that is the problem. The whole logic for this would have to be implemented. Think switches, commands to the linker, a way to configure the name for the segment etc. Part of this is specific for your device setup.

If writable const is off, also variables declared with CONST could be put in a different segment. There is no need to invent new syntax.

However all the linker stuff should be implemented by sb who has such a device and the interest to work on it.

Why not just drop static constants in the Text (Code) segment. I have seen other compilers do that. This will not involve any change of the interface with the linker (That is what I am doing now in a separate assembler file).
On the other hand I am already using a configuration file for the linker. (Where to place Stack, data code etc.) I assume that it would be easy to take care of an extra segment in the linker configuration file.

Regards Carsten

I might be completely wrong, but I suspect it isn't such an easy decision (as a general rule at least). Some CPUs/platforms might not allow data access in TEXT (code) don't know whether this is the case for any of our currently supported platforms (I surely know that it is _not_ the case for x86 at least ;-) and I guess the same is true for ARM if you're doing it right now), but it might be (either now or in future).

Tomas

In the compiler is support for a .rodata section, but this doesn't work yet. It has very low priority to fix this support. You are welcome to supply patches to make it work. In your own linker script you can then include the .rodata in the .text segment of the final executable.