This is the mail archive of the
libffi-discuss@sourceware.org
mailing list for the libffi project.
Important LIBFFI bug in PowerPC subtree
- From: Radu Sion <sion at cs dot sunysb dot edu>
- To: green at redhat dot com
- Cc: libffi-discuss at sources dot redhat dot com
- Date: Thu, 15 Jul 2010 16:23:15 -0400 (EDT)
- Subject: Important LIBFFI bug in PowerPC subtree
Anthony, all,
there seems to be a somewhat important bug in the PowerPC subtree of
libffi that makes it unusable on powerpc chips with no floating point support.
To summarize: despite being told no floating point hardware
exists at configure/compile time, libffi-3.0.9/src/powerpc/sysv.S
contains active floating point instructions for loading/unloading
floating point values when invoking functions.
E.g., lines 88-96, etc.
This results in "Illegal Instruction" being thrown by the target platform
(PowerPC 405) -- and as no kernel-emulation of floating point exists, this
crashed anything linked with libffi.
It is unclear how to change the calling convention to match whatever
gcc would generate for this, but as it stands, FFI cannot be used to
invoke any floating-point related code on hardware without floating
point support if it returns a float or has a float parameter.
cheers,
radu
--
radu sion
http://www.cs.stonybrook.edu/~sion