Re: glibc patch for TLS?

[Michael Long <nikarul@gmail.com>]

That would be helpful, though the access still could be inside a
library, which would be a be pain to track down in say, Mozilla or
KDE.  Still, it's better than nothing.

  -Michael

On 4/25/05, Ian Pratt <m+Ian.Pratt@cl.cam.ac.uk> wrote:
> 
> 
> > Nope, that's what I forgot.  I recompiled the patched glibc
> > with that flag and everything seems to be working fine.  I
> > can post the ebuild if people are interested.  It's pretty
> > straightforward, just adding the patch and making for the
> > CFLAGS get set properly.
> >
> > I'm now compiling all my packages with
> > -mno-tls-direct-seg-refs.  Is this necessary/recommended?  I
> > know some ebuild strip out CFLAGS it's unsure of, so if it's
> > necessary other packages may need to be modified.
> 
> Glibc really hammers the thread local storage stuff and hence
> experiences rather more of a slow down rather than any applications
> we've come across, but it would obviously be preferable to recompile
> anything that uses tls.
> 
> One of the hard things is figuring out which applications actually make
> use of the thread local attribute, and hence would actually benefit from
> recompiling.
> 
> I guess we could modify the warning message to print out the name of the
> process that did the -ve segment access...
> 
> Ian
>