NPTL Merge Status.

[Peter S. Mazinger]

On Tue, 22 Aug 2006, Erik Andersen wrote:

> On Tue Aug 22, 2006 at 12:48:42PM +0200, Peter S. Mazinger wrote:
> > I would prefer to see at least the non-new-linuxthreads/non-new-features 
> > parts from my megapatch in, else .29 will be released buggy.
> 
> Could you extract said portions of your megapatch and post them
> for review?  As discussed, I'm not going to wade through a
> megapatch that does 50 separate things.  If you want to get
> patches in, a sequence of patches vs trunk adding one feature per
> patch is the way to go.  Your email from the other day noting a
> couple of superfluous headers was a fine example of how to do
> things right, and resulted in immediate action.  Anyway, now
> would be a great time for you to try and get things into trunk,
> since I actually have time this week to work on uClibc...
> 
>  -Erik
> 
> --
> Erik B. Andersen             http://codepoet-consulting.com/
> --This message was written using 73% post-consumer electrons--

Began parts of it, I have 1-2 days to do it, after that I won't have time 
for anything uClibc related and the rest will have to be done by someone 
else needing working new-linuxthreads. I will try to get those that are 
also affected by NPTL.

proposal:
- add UCLIBC_USE_TLS config option that is enforced by NPTL and is 
optional for new linuxthreads.
- provide generic tls.h that can be included anytime (independently of 
uClibc configs) gets uClibc config options and depending on UCLIBC_USE_TLS 
defines USE_TLS and USE__THREAD (as used in glibc and nptl branch).
- enabled UCLIBC_USE_TLS should bail out on improper __thread support in 
compiler (currently done in Rules.mak for NPTL). If USE_TLS is defined, 
include_next arch_specific tls.h (or we could move the glibc_arch_specific 
tls.h to arch/bits/tls.h and include that, no need for include_next then).
- provide generic sysdep-cancel.h/not-cancel.h or other nptl needed 
headers and again handle there the case if they should do anything or not 
depending on config options. The benefit is that we do not have to guard 
for cases if they are not needed and new-linuxthreads needs most of them 
too, so later ifdef UCLIBC_HAS_THREADS_NATIVE has to be changed to if 
defined UCLIBC_HAS_THREADS && !defined LINUXTHREADS_OLD.
- see also my comments on arm-nptl patch please

Peter 

-- 
Peter S. Mazinger <ps dot m at gmx dot net>           ID: 0xA5F059F2
Key fingerprint = 92A4 31E1 56BC 3D5A 2D08  BB6E C389 975E A5F0 59F2