AnsweredAssumed Answered

Forward porting custom blackfin kernel modifications

Question asked by stamppot on Mar 27, 2013
Latest reply on Mar 28, 2013 by Aaronwu



I'm currently investigating a series of problems with a BF526-stamp used in a product. We're adding WiFi to Ethernet connectivity via USB, but we're having serious bandwidth issues, caused by enormous amounts of packet drops. Part of our investigation is migrating to the latest buildroot-based distribution, and at the same time untangling some kernel problems. This is where our current problem occurs:


I've derived a kernel tree with minimal modifications for our specific board, based on the BF526-stamp configuration. When compiling a buildroot with this kernel, the kernel compile fails with large amounts of 'cannot link non-fdpic object file into fdpic executable' errors, which of course result in linker errors down the line. When selecting the default BF526 stamp configuration in the kernel menuconfig, everything compiles just fine. I have verified both defconfigs and they are almost identical (apart from our addition of the custom board, of course) and are identical in terms of executable format support, etc. Please see the attached log for the full compile logging. I have also attached our current defconfig.


Furthermore, if I then change into the output/build/linux-custom directory and execute a make there (just for testing purposes), none of these link-errors occur! The compile will fail at the end when linking the kernel, but that's another matter. I am at wit's end trying to find out what is wrong here. Can anyone shed some light here?


Further information:

  • Build is done using the buildroot-2012R2-RC5
  • Toolchain used is toolchain-2012R2-RC5_x64
  • All compiles are executed after doing a full erase of the output/build directory. I have been bitten by dangling compiles too often...

If you need any additional information, please let me know.