Re: Re: pthreads in Linux

[Kapish K <kapish at ureach dot com>]

Hello,
	Well, I looked at the code, and there are a few queries on
this:
Firstly, yes, sysdeps/i386/i686, there is a pt-machine.h, which
includes useldt.h ( which has the #define for the register
support ) and pt-machine.h is included in internals.h, which is
included when linuxthreads is compiled.
But, however, note that the useldt.h include in pt-machine.h
itself is commented out with the following comment:
/* Use the LDT implementation only if the kernel is fixed.  */
//#include "../useldt.h"   

Now I am confused.. What does this 'kernel is fixed' supposed to
mean? Does this get included or not for i686 and why not for
ix86 other than i686?
Any pointers?

TIA



---- On Wed, 9 May 2001, Wolfram Gloger
(Wolfram.Gloger@dent.med.uni-muenchen.de) wrote:

> > In the code, I
> > saw that for non-standard stacks, in thread_self,
> > it just parses through the whole list of pthread
> > stacks to match one with the current sp.
> 
> Yes, I believe this is the case, but only if the thread
descriptor
> can't be retrieved by other means.  If glibc is compiled for
i686 on
> recent Linux kernels for example, a specific register is used
to hold
> the thread descriptor, and the list of stacks is _not_
traversed.
> 
> > 'coz we
> > are concerned this might causes performance
> > problems for us,
> 
> Really?  Why do you need user-defined stacks, and so many of
them?
> IMHO there is very rarely a justification for user-defined
stacks, and
> the POSIX semantics are not very friendly..  In particular,
you cannot
> safely re-use the memory areas for thread stacks, because
`thread
> termination' isn't well defined.
> 
> Regards,
> Wolfram.
> 
> 


________________________________________________
Get your own "800" number
Voicemail, fax, email, and a lot more
http://www.ureach.com/reg/tag