Re: striking out at the eabi

[Mike Taht <mike at ackley dot net>]

On Sun, 7 May 2006, Michael K. Edwards wrote:

> The "linux headers are too old" message is an artifact of the way that
> crosstool bootstraps linux headers into the first stage of the glibc
> preparation using the host's gcc.  There are changes in syscall
> numbering between the old ARM ABI and the EABI which only happen if
> __ARM_EABI__ is defined.  Your x86 host gcc of course doesn't define
> this symbol, so the glibc configure step errors out when it detects
> the old syscall numbering.
> 
> The patch attached to "More crosstool-0.42-glibc-2.4-gcc-4.1.0-nptl"
> works around this by adding a "GLIBC_FAKE_CROSS_ARGS" parameter to
> crosstool.  The value of this parameter gets tacked onto commands that
> pretend that the host gcc is the target gcc; setting it to
> -D__ARM_EABI__ (as in my earm.dat) resolves the "old headers" problem.

I'd pulled down the wrong or an obsolete patch. I don't remember 
where I got "crosstool-0.42-mg2.patch" but that did not have any faking in 
there.

Rebuilding now. I do look  forward to having a standardized softfloat 
userspace! 20 years of pointless hardware traps on floating point....

ntpl, too. Joy. Bliss. Rapture.

> I would be interested in hearing what results you get with that patch,

Coming up. One nice thing about doing these x86_64 builds - I can do a 
couple dozen in a day - not like, one, overnight, on my soontoberetired 
rh9 box....

> especially if you will be using binutils 2.16.92.  I am still working
> on kernel 2.6.16.11, having ported the RT/preempt patch to it; once I
> work through the rest of the arch/mach issues for my board, I may move

cool, you mean genericarmirq?

> on to 2.6.17.  Note, however, that apparently Linus has announced that
> 2.6.16 will be supported (i.e., security and critical-bug-fix patched)
> long-term, so it may be a good branch point for embedded developers.

I felt the new mutex stuff needed some shaking out in .17, and 
only half the ep93xx support only just went into .17, so I think I'll be 
tracking the .18 and .19 kernels... and then there is seeing the 
genericirq stuff get folded in... maybe .20 will be the magic stable 
number! we can always hope.

> 
> Cheers,
> - Michael
> 

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