Re: uClibc and pthreads

Hi Pascal,

On 2024-01-24 12:53, Pascal Scholz wrote:
> 1) Is there a technical reason for L4Re still using the Linuxthreads
> implementation and not some kind of NPTL port or other
> implementations?

The libpthread implementation used in L4Re is based on the linuxthreads 
(formerly named linuxthreads.old) implementation from uClibc. There is 
no particular reason why one was chosen over the other that I am aware 
of. In the end it doesn't matter that much, since the "backend" is 
highly L4 specific anyways. I guess the linuxthreads implementation was 
easier to port.


> 2) What uClibc is used exactly? Do you manage your own fork of uClibc
> and back port relevant patches of uClibc-ng? Or do you use a uClibc-ng
> with additional L4 patches? Or some completely other version? It seems
> that the uClibc version used diverged in 2014 from the original. This
> seems relevant to me as I wanted to check, which changes were done to
> make the uClibc work with L4 (in an attempt to generate some kind of a
> diff), so I might port them to another libc implementation.

Originally uClibc was used and adapted to L4Re. When uClibc was no 
longer actively maintained we implemented some of the things we needed 
ourselves or ported them from other C libraries. After uclibc-ng gained 
traction we started first porting fixes and improvements from them when 
we needed them. More recently we started to synchronize uClibc with the 
current upstream uClibc-ng state and also already contributed some of 
our changes back where this made sense to us. This is not yet done but a 
work in progress and you will likely see more changes to our uClibc 
during the next months.

> In my project I would like to replace Linuxthreads with some other
> pthreads implementation and measure the performance impacts of these
> changes, so the aforementioned questions seem relevant to me.

The best start is probably to to look at the libpthread implementation 
you find in l4re-core/uclibc/lib/libpthread and compare this to the 
upstream/uclibc-ng version of these files. Though the upstream might 
have diverged in the meantime. We only imported changes when they were 
known to be performance critical or bug fixes but otherwise kept our 
implementation stable so far. This might as well change once we caught 
up with upstream uclibc-ng.

I would be very interested in the results of your measurements! Do you 
already have other libpthread implementations in mind that you want to 
look at? As noted in the BENCHMARKING file in l4re 
(https://github.com/kernkonzept/mk/blob/master/BENCHMARKING) a heads up 
to [email protected] on any measurements before they are 
published would be highly appreciated so that we can sanity-check them 
and if they match our experience with the performance we see.

If you have any questions while looking at the code don't hesitate to 
ask.

Best regards and happy hacking!

- Marcus

> Thanks in advance for you time!
> 
> Kind regards,
> 
> Pascal
> 
> _______________________________________________
> l4-hackers mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
_______________________________________________
l4-hackers mailing list -- [email protected]
To unsubscribe send an email to [email protected]