On (19/09/10 19:18), R.T. McGee wrote: > The Faraday FA526 core is used in ARM processors like the Semi STR8132 > / Cavium CNS2132 which is in my Technologic Systems TS-7500 board in > addition to some NAS boxes out there. From what I have read, it > basically resembles a ARM920T but has NO thumb interworking support > which makes it not fully EABI compliant ( > http://www.mail-archive.com/[email protected]/msg05856.html > ). I have build a few tool chains manually that produce working code > using a configuration meant for a OABI style armv4 which makes sense > since it uses NO thumb instructions, but I am not sure this is > optimal.
eabi is now supported on armv4 architecture and we have OE internal toolchain also support it from gcc 4.4.x onwards. > > What is the best way to move forward to setup a OpenEmbedded > distro/machine configuration for something like the TS-7500 or a > general FA526 configuration? look at any armv4 machines we have in OE e.g. collie and distros like minimal or angstrom DISTRO=minimal MACHINE=collie bitbake minimal-image I guess, one option is setup OABI ARMv4 > style setup but this was removed from things like angstrom a while > ago. Switching to OABI just to get rid of thumb seems a little silly. > > It would be nice to figure out a way to use EABI ARMv4T toolkit but > pass flags like -mapcs-32 -mno-thumb-interwork (just a guess) to GCC > to insure no thumb instructions (other weird things are discussed in > http://wiki.debian.org/ArmEabiPort#Otherscenarios ). The linker > option -fix-v4bx looks promising for the linker also. A lot of things > are possible solutions, but whats the cleanest way to work it into > OpenEmbedded's build system? its already functional in OE. > > Any suggestion would be helpful. Thanks. > Thx -Khem _______________________________________________ Openembedded-devel mailing list [email protected] http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
