[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+qPFcK2Trv7qCVy-FnsCw0nEdsWn=5pY+rpz1-vAzdQAobwJQ@mail.gmail.com>
Date: Tue, 02 May 2017 15:49:36 +0000
From: mzpqnxow <musl@...qnxow.com>
To: musl@...ts.openwall.com
Subject: Re: Are there plans to support the "old" ABI (apcs-gnu) for ARM?
I suppose what really kills me is the different system calling convention
on Linux 2.4 (which in my case seems a hard requirement) not just the
instruction set- but I realize now that musl does not aim to support 2.4
(which is perfectly reasonable, same as not supporting the ARM OABI- total
obsolescence)
Thanks again for the details and the ld flags, I wasn't familiar with them,
I might be able to hack around things between the linker trick and a
(pretty involved) binary patching or patches to musl I can whip up- lots of
fun!
I'll GPL whatever I come up with in case anyone else is interested but I
realize it won't go into musl upstream without someone to support it. I
don't have the time or expertise for bugfixes and/or QA, so that won't
happen in sure.
Unfortunately in my case I'm stuck on Linux 2.4 and ARM OABI for this
"project" :/
I wonder if there's an easy way to work around requiring the NSS libs by
just patching uClibc.. I was able to build what I needed with an older GCC
and uClibc, aside from the NSS libs static linking warnings (sorry, a
little off topic now)
On Mon, May 1, 2017 at 18:45 Rich Felker <dalias@...c.org> wrote:
> On Tue, May 02, 2017 at 12:05:51AM +0200, Szabolcs Nagy wrote:
> > * mzpqnxow <musl@...qnxow.com> [2017-05-01 16:11:10 -0400]:
> > > I've seen it documented in several places that the so-called "old" ABI
> for
> > > ARM (formally referred to as apcs-gnu, I believe) is not supported by
> musl,
> > > though I haven't seen any formal explanation as to why. I'm wondering
> if
> > > this is a design decision that has already been made or if it is just
> very
> > > low priority on the TODO list- but might be implemented some day. I
> have a
> > > need for an ARM OABI toolchain that allows me to statically link a good
> > > amount of non-trivial executables and uClibc, glibc, etc. don't work
> for
> > > me, mainly due to the reliance upon functions in the nss libraries
> which
> > > need dynamic loading to function correctly.
> > >
> > > So I'm curious, will this "old" ARM ABI ever be supported/implemented
> in
> > > musl or is it dead enough "in the wild" that it will never be
> implemented?
> > > Also, if there is some way that I am missing something and the old ABI
> *is*
> > > supported, I'd be happy for someone on the list to point out how I
> missed
> > > it.
> > >
> > > A simple yes/no would be terrific, I don't need a long explanation of
> how
> > > obsolete and slow the old ABI is- I am painfully aware. But here I am
> :>
> >
> > oabi would be a completely new port (even the syscall abi
> > and type representations are different, not just the pcs)
> >
> > and it would be more work than a normal port because oabi
> > has mixed endian double representation and the math code
> > is not prepared for that (most double prec math code would
> > need some fix).
> >
> > it's unlikely to be supported any time soon
>
> I would add that it's unlikely to be accepted upstream since it's
> obsolete and doesn't even have any maintained compilers that can
> produce code for it (gcc dropped it years ago). If you need pre-armv4t
> support, the right solution would be using -Wl,--fix-v4bx or maybe
> -Wl,--fix-v4bx-interworking.
>
> I'm not sure if there's an easy way to get old oabi kernels to run
> such code; it's probably not too bad if you have the kernel source and
> can rebuild, but may be quite hard if you can't easily replace the
> kernel.
>
> Rich
>
Content of type "text/html" skipped
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
