Re: Alignment problem with PPC cross assembler (gcc code)

[Alan Modra <amodra at bigpond dot net dot au>]

On Tue, Jun 17, 2003 at 06:34:57PM +0100, Trevor Marshall wrote:
>    3              		.csect .text[PR]
>    4              		.csect _testtask.rw_c[RW],3
>    5              	bad_var.0:
>    6 0004 23       		.byte	35

OK, at this point we're in a data subsection and currently at address 5.

>    7              		.csect .text[PR]

Now we've changed to a text subsection.

>    8              		.align 2

And align the text subsection.

>    9              		.globl bad_func
>   10              		.globl .bad_func
>   11              		.csect bad_func[DS]

Back to a data subsection, a different one.  This one happens to follow
after the _testtask.rw_c[RW] subsection, which doesn't have any further
contents.  So this subsection starts at address 5.

>   12              	bad_func:

Set a label in the new subsection.  At this point there hasn't been an
align, so the label is at address 5.

>   13 0005 00000000 		.long .bad_func, TOC[tc0], 0

Start emitting to the data subsection.  .long happens to align before
emitting, so the first element, .bad_func, gets stored at address 8.

I'm not sure what other assemblers do with a .csect directive.  If your
testcase assembles correctly on AIX, then it looks like gas should be
aligning at the start of each subsection.  Could you try the testcase
under AIX?

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre