Re: ARM support for glibc-2.5

["Michael K. Edwards" <medwards dot linux at gmail dot com>]

On 3/3/07, Robert Schwebel <r.schwebel@pengutronix.de> wrote:
> We don't use the crosstool patch stack for our toolchains, at least not
> exactly. Please be patient until monday, we'll release our toolchain
> build system then.

Ah, I understand.  I'd be delighted to see ptxdist's toolchain
construction support a soft-float ARM OABI target.  (If someone's
feeling industrious they could adjust it to use VFP rather than FPA,
for a cleaner transition to EABI.)

One thing to be aware of: the gcc / glibc / glibc-ports /
linux-headers dance is somewhat involved, and the trick with which I
am familiar actually uses the host compiler for one of the
preprocessor passes over the Linux headers.  That will not necessarily
work correctly if you don't define some extra preprocessor symbols.  I
ran across this when building an EABI toolchain (if you don't define
__ARM_EABI__ you get the wrong syscall numbers), and that led me to
add the GLIBC_FAKE_CROSS_ARGS mechanism to crosstool.sh.  But there
may be other flags hiding inside crosstool.sh that were already taken
care of before I started working on it; in particular, I started from
some work that Martin Guy had done to fix the building of NTPL
toolchains.

What I would suggest that you do is to build a toolchain with
crosstool first, just to make sure that I haven't left out any pieces.
If you hit more snags I'll repost my crosstool changes as a patch
(maybe I'll port forward to crosstool 0.44 and retest with gcc 4.1.2
while I'm at it).  Then it'll be easier to compare the logs against
your build system's logs and see where things go wrong.

Cheers,
- Michael

--
For unsubscribe information see http://sourceware.org/lists.html#faq