This is the mail archive of the binutils@sources.redhat.com mailing list for the binutils project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Problem with VFP indication in ARM objects


Hi Jason,

> NetBSD/arm ELF uses VFP-format soft-FP, and modern gas can mark objects
> as such...
> 
> However, the 2.11.2 gas could not mark objects as such...
> 
> "Guess which version of gas is currently used in NetBSD."
> 
> I need to come up with a solution which allows me to link objects which
> are properly marked as "soft-VFP" with objects which are not (the objects
> in question unfortunately lack both flags, so appear as "hard-FPA").
> 
> Anyone have suggestions of how I can lessen my pain?  Am I stuck with
> a flag-day?

We could remove the test/error message from the BFD library.  The test
is in there in order to catch the accidental linking of incompatible
binaries, but if this test is going to prevent the linking of
compatible binaries then it can be considered to be flawed.  This
could be controlled by a command line switch, and limited to just the
"presence" of the hard-FPA flag.

We could have the code in BFD detect the absence of any flags, and
then scan any code sections to see if any hardware VFP instructions
actually exist.  If they do not, it could remark the binary as
soft-VFP.  Ugly, but doable.

You could write a utility to change the flags in an ARM ELF binary.
You could then use this tool on your current binaries to set the
soft-VFP flag.

Or, as Richard suggests, we could switch over to having a .note
section with proper information in it.  You could even propose such a
section to the ARM EABI committee which is currently discussing how to
improve/fix the ARM ABI.

Cheers
        Nick




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]