Re: [RFC] dietlibc on hppa: thread local storage setup
On 03.04.2016 17:28, Christian Seiler wrote:
> Dear hppa porters,
> I've recently started helping out with the dietlibc package. Regarding
> hppa I've merged all porting patches in the bugtracker.
Thanks for doing that!
> However, an
> upload to experimental based on a current CVS snapshot didn't
> immediately build: the new package (and upstream) unconditionally
> enable TLS support on all platforms and define the errno variable as
> thread-local via __thread. This means that every platform needs to have
> TLS setup code present (before main is run), otherwise any access to
> errno (i.e. any system call) will fail.
> I've looked at both the glibc and kernel sources plus read a lot of
> documentation about PA-RISC and I think I've figured out what to do.
> The latest upload to experimental now builds on hppa (and some tests
> are run during build process, so at least simple programs work), so I'm
> quite confident that this is correct.
> My change can be found in this patch:
> (It also addresses TLS startup for other archs, please ignore that.)
This looks good.
> Since I don't have access to hppa hardware myself, qemu-user doesn't
> seem to support it and there aren't any porterboxes - at least to the
> best of my knowledge - I would be grateful if you could take a look and
> see if what I've done is correct. I want to send all of the patches
> w.r.t. porting currently in Debian to upstream, but I'd really like a
> second opinion on this change.
Right, we don't have porterboxes and qemu doesn't support hppa (yet).
But I'm happy to give you a login on one of the hppa buildd servers
if you like.
> (Also, if there are other issues with dietlibc on hppa that you haven't
> reported, feel free to mention them. Please look at the most recent
> version in experimental, 0.34~cvs20160402.1-1~exp2, though.)