GCC440 (arm) : (.data) section is empty in output

May 13, 2009 5 Replies
F

Good day, I've noticed that the initialized data section in the program binary is zero bytes, by inspecting the file with 'size':


$ arm-none-eabi-size main.out text data bss dec hex filename 56160 0 2856 59016 e688 main.out



I'm running GCC 4.4.0, crosscompiled for ARM under Linux. I definately have the ".data" section in my linker script. The data section follows my (.text) section.



I've confirmed that I have some variables that are initialised inside my code where they are defined, i.e


---snip--- int main(void) { char myChar=5; ... }



Could somebody please give me some feedback on the way that GCC is handling initialized variables and if what I'm seeing is actually the correct behavior?


---snip--- .data 0x04000000 : AT (ADDR(.text) + SIZEOF(.text)) { . = ALIGN(4); datastart = .; __data_start__ = . ; *(.data) . = ALIGN(4); __data_end__ = . ; edata = .; _edata = .; }



---snip---



Regards, Frikkie Thirion



Vote
Twitter Facebook LinkedIn
N

Since myChar is an automatic variable (declared within a function and without the "static" keyword), it is placed in the stack and not in a statically allocated memory location. The variable is initialized dynamically by code that is executed whenever "main" is entered and its stack frame is set up. Therefore I would not expect this variable to add anything to the "data" section.

Niklas Holsti Tidorum Ltd niklas holsti tidorum fi . @ .
Vote
Twitter Facebook LinkedIn
H

It is correct. Things were different when you would have declared myChar as being static. Only in this case the room occupied by char is permanent and can be statically initialized (i.e. at program starting time through .data).

Heike

Vote
Twitter Facebook LinkedIn
F

Good day, Thank you for the replies. I've done the following tests: 1) declared myChar static --> The size command now shows that .data occupies 4 bytes.

2) I've searched through the project and found a couple of other variables that are declared statically, but initialised as 0. The compiler/linker then decided not to add them to .data, but apparently to .bss.

As soon as I've changed some of them to be initialised to something other than 0, they appear in the .data section. Seems to be a good optimization, since the general practice for .bss (uninitialised data) is to clear it at startup time.

Vote
Twitter Facebook LinkedIn
H

That's not just "general practice" --- it's a requirement of the programming language. Some tools will let you turn that off (it could even happen by accident; BEWARE!), but by doing so you leave solid ground. Among other things, it opens up your program to bugs that hardly any code testing method in the world can help you discover, since they'll invariably assume that the tools behave like the language says they should.

Vote
Twitter Facebook LinkedIn
N

No it isn't. The C standard doesn't say anything about segments or executable formats.

Static variables lacking an explicit initialiser are implicitly initialised to zero (rather than containing garbage like uninitialised automatic variables), but this doesn't have to use the "bss optimisation"; the compiler is free to store all static variables in the same way, regardless of whether they are zero.

Vote
Twitter Facebook LinkedIn

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required