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: [PATCH/RFA] Mark arm-*-netbsdelf* binaries as ELFOSABI_NETBSD


> On Sun, 2002-04-07 at 07:09, Jason R Thorpe wrote:
> > The upcoming NetBSD 1.6 release is shipping with ARM ELF binaries marked
> > with ELFOSABI_NETBSD, in order to identify the various characteristics of
> > the ABI we are currently using (ATPCS, VFP-format doubles, but non-packed
> > enums).
> 
> That is not really what the EI_OSABI field is for.  It is intended to
> identify the ABI in a wider sense, so that you can ascribe meanings to
> things like reloc numbers.  If your object files are compatible with
> what elfarm-nabi.c produces, you should stick with ELFOSABI_ARM.

For a fully conforming ARM EABI implementation, this information will be 
determined by looking at EF_ARM_EABI_VERSION (the top 8 bits of the 
EF_FLAGS field).  So there's no need to mandate the setting of the 
ELFOSABI field.

BTW, who allocated ELFOSABI_ARM?  I'm fairly certain it wouldn't have been 
ARM, since the existing ARM ELF documents (based on TIS ELF) don't go as 
high as that field (saying that it should be left at zero).

> 
> What were you planning to use this identification for?  Is it just to
> stop people from accidentally linking together incompatible objects, or
> is there something else?

It's mainly intended as a run-time execution identifier.  But it's also 
useful for identifying incompatible objects when we finally switch to the 
full ABI.

R.



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