On Tue, Jun 10, 2003 at 19:38:56 -0700, Jason Thorpe wrote:

> >The problem is that the kernel ld script for netwinder does not list 
> >link_set_* sections, so they are transfered to the ld output as-is 
> >after the .text section, however the kernel ld script explicitely 
> >places .data section after the .text section with:
> 
> I have certainly not seen this problem, and I know I've built/booted
> netwinder kernels recently.  The linker should actually be sorting
> the link_set_ sections along with the .rodata section.

Should it?  My readind of ld(1) docs is that the "orphaned" sections
are transferred from input to output as-is.


> >  .data    :
> >  AT ((LOADADDR(.text) + SIZEOF(.text) + (0x8000 - 1)) & ~(0x8000 - 1))
> >  { ... }
> >
> >if sizes of .text section and link_set_* section are such that there 
> >is not enough room for the link_set_* sections in the last page 
> >occupied by the .text section, the link_set_* section will be 
> >overlapped by the .data section b/c of the explicit .data section 
> >placement
> 
> ...oh, right, SIZEOF().  "Nice."  This is a problem with the use of 
> SIZEOF(); it cannot be safely used when orphan sections (like link_set) 
> are in use.  I submitted a patch to "fix" it to the FSF, but it was 
> rejected, so I implemented a different strategy for the evbarm port...
> 
> Take a look at evbarm/conf/ldscript.evbarm -- it should have the hints 
> you need to fix the netwinder script.

Oh, that helps.    Is that a correct change:

   . = ALIGN(0x8000);
   .data    :
-  AT ((LOADADDR(.text) + SIZEOF(.text) + (0x8000 - 1)) & ~(0x8000 - 1))
+  AT (LOADADDR(.text) + (ADDR(.data) - ADDR(.text)))

I think the preceding ALIGN is enough to avoid the bit-masking dance
of the old version (and the resulting kernel correctly places the
.data), but just to be sure...

Thanks.

SY, Uwe
-- 
uwe@ptc.spbu.ru                         |       Zu Grunde kommen
http://www.ptc.spbu.ru/~uwe/            |       Ist zu Grunde gehen