ld fails with .bss / .rel.dyn overlap for 2011.12

This email is to report a problem building u-boot 2011.12 with a gcc 4.3.2 toolchain. I've added support locally for a marvell 78xx0 based board. It's using arm926ejs and I have not modified the *.lds linker script.

Case 1: With gcc 4.3.2 provided by marvell, the build fails with this error:

arm-mv5vfp-linux-gnueabi-ld: section .bss [00000000fffbba08 -> 00000000fffc3137] overlaps section .rel.dyn [00000000fffbba08 -> 00000000fffbeea7]
arm-mv5vfp-linux-gnueabi-ld: section .dynsym [00000000fffbeea8 -> 00000000fffbef57] overlaps section .bss [00000000fffbba08 -> 00000000fffc3137]
arm-mv5vfp-linux-gnueabi-ld: u-boot: section .bss lma 0xfffbba08 overlaps previous sections
make: *** [u-boot] Error 1

Through trial-and-error, I found that this error only occurs when .bss is unaligned. In my example .bss is only 2-byte aligned and u-boot.map indicates that the linker added 2-bytes of zero fill at the end of bss. If I add a global unsigned short so .bss is 4-byte aligned the error goes away and gcc produces a good binary.

Case 2: With gcc 4.6.3 built with crosstool-ng, the build has no problem.

I can provide u-boot.map files (~70k) for each case.

-joey

Dear Joey Oravec,

...

Thanks, but what should we do about that? You are referring to code that is unknown to us, and to a tool chain we have no access to.

It would probably be best to report this to the Marvell compiler team.

Best regards,

Wolfgang Denk

DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
There were meetings. There were always meetings. And they were  dull,
which is part of the reason they were meetings. Dull likes company.
                                    - Terry Pratchett, _Making_Money_

Possibly nothing -- I'm reporting it because there have been other mailing list thread with the same error message, sometimes without a clear resolution:

http://lists.denx.de/pipermail/u-boot/2012-February/117009.html
http://lists.denx.de/pipermail/u-boot/2010-November/082122.html

I don't know enough about the linker or *.map file to understand the root cause of the error message. I can only report my observations and suggest that if somebody else hits the same problem they should try using a newer toolchain. In my case it fixed the problem.

I don't need anybody to take any action on this, so feel free to disregard if this report is not interesting. I have no support through Marvell but I thought it was appropriate to share my results with this mailing list.

-joey

Dear Joey,

Please don't misunderstand me. This is certainly interesting information, and probably even important to some. Unfortunately we can do nothing to help it, as we don;t have access to the Marvell code nor tool chain.

I think you should still report it to Marvell, so they know as well.

Best regards,

Wolfgang Denk

DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Drun'? 'm not drun'! You woudn' dare call m' drun' if I was sober!
                                     - Terry Pratchett, _Men at Arms_

If I read the logs correctly, this was seen with other tool chains, e.g. Codesourcery ones, too.

But if I remember correctly, the conclusion was to use newer, not 'broken' tool chains. Whatever 'broken' does mean here ;)

Best regards

Dirk