Re: More ARM binutils fuckage

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

Thank you again Lennert, this is really helpful.  As you probably
expected, I have a few more questions.  :-)

On 12/5/06, Lennert Buytenhek <buytenh@wantstofly.org> wrote:
> > Do you turn on TLS and/or OABI userland compatibility in your kernel
> > config?
>
> What is TLS userland compatibility?

Sorry, ambiguous phrasing.  I was asking about
 1) TLS;
 2) OABI userland compatibility.
What I really meant to ask about TLS is whether your processor has the
new CP15 "hard" TLS register (presumably only accessible from an EABI
userland binary) or whether perhaps you are using "soft" TLS, which
may be usable with an OABI binary and an appropriately configured NPTL
glibc.

> Yes, I build kernels with OABI compatibility, as I said above.

Do you happen to know whether this works properly in 2.6.16.x?
2.6.18.x?  2.6.19 with Ingo's preempt/RT patch?

> > Is this toolchain built with crosstool,
>
> Nope.  I made use of the excellent work done by the OpenEmbedded
> people (and more specifically Koen Kooi) on the EABI Angstrom distro,
> and from there I natively built binutils/gcc/glibc on the actual
> target (so the EABI compiler and C library I use on the ARM board
> were actually compiled on that same ARM board.)

Have you tried running your native toolchain under qemu?  That's been
a successful strategy for me in the past, especially since a lot of
configure scripts don't get cross-compile quite right.  I've also been
accumulating a set of patches to PTXdist for cross-compiling an EABI
userland, which I'll split into manageable fragments and post over
there once I have actually put it through its paces on my hardware.

> > Does NPTL work for you in this configuration?
>
> Yes.

You wouldn't happen to have benchmarked a thread-intensive load on
your hardware with and without NPTL, would you?  I would expect the
gain to be significant from not blowing MMU context on every thread
switch, but I haven't seen hard numbers on ARM.

Cheers,
- Michael

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