Re: [Buildroot] [autobuild.buildroot.net] Daily results for 2022-09-22

From: Waldemar Brodkorb <wbx@openadk.org>
To: yann.morin@orange.com
Cc: Stafford Horne <shorne@gmail.com>, buildroot@buildroot.org
Subject: Re: [Buildroot] [autobuild.buildroot.net] Daily results for 2022-09-22
Date: Fri, 23 Sep 2022 17:23:11 +0200	[thread overview]
Message-ID: <Yy3PX2eJUrQWvxw+@waldemar-brodkorb.de> (raw)
In-Reply-To: <10091_1663942719_632DC03F_10091_295_1_20220923141837.GD2981@tl-lnx-nyma7486>

Hi Yann,
yann.morin@orange.com wrote,

> Waldemar, All,
> 
> On 2022-09-23 14:13 +0200, MORIN Yann INNOV/IT-S spake thusly:
> > On 2022-09-23 05:30 +0000, Thomas Petazzoni via buildroot spake thusly:
> > >     or1k     |           gpsd-3.24            | NOK | http://autobuild.buildroot.net/results/7adc125e843b21b559f1e8813059d65af58feb8d |     
> > 
> > This is an or1k, shared-only, uClibc-ng, LT (not NPTL) so no TLS, build.
> > 
> > So, I'm not sure to make of this issue:
> > 
> >     ..../or1k-buildroot-linux-uclibc-gcc -o gpsd-3.24/gpsctl .... -lusb-1.0 -lm -lrt -lnsl
> >     ..../ld: ..../sysroot/usr/lib/libusb-1.0.so: undefined reference to `__tls_get_addr'
> 
> So, here are a few other tests I did (each independtly):
> 
>   - revert uclibc-ng from 1.0.42 back to 1.0.41 -> still fails
>   - revert libusb from 1.0.26 back to 1.0.25 -> still fails
>   - revert gpsd from 3.24 back to 3.23.1 -> still fails
> 
> So, from the look of it, the __tls_get_addr has alwaus been injected in
> libusb [0] on or1k with uclibc-ng and linuxthreeads.
> 
> Now, is that a valid combo?

Normally I would say it is. May be this is an gcc/binutils issue?
Maybe the uClibc-ng or1k maintainer knows more? Stafford do you have
a plausible explanation for this phenomenon?

> Do we want to keep LT even for those arch where NPTL exists? I.e. when
> uclibc is configured with threads, use NPTL is there is an NPTL
> implementation for the arch, otherwise use LT. Thoughts?

It might be a solution, yes. But do not forget it is not that
simple, as it also depends on the binary format used, as for ARM
where pending patches for FDPIC will use NPTL and FLAT is
Linuxthreads only. Sometimes it might be useful to use Linuxthreads
to see if it is a NPTL bug or not. May be Stafford have an idea.

best regards
 Waldemar

> [0] I don't think it is specific to libusb, just that this is the first
> time it got noticed and investigated...
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot