[lowrisc-dev] What do you use Debian for (in lowrisc)?

Aurelien Jarno aurel32 at debian.org
Fri Dec 28 12:16:46 GMT 2018


Hi,

On 2018-12-28 12:55, Manuel A. Fernandez Montecelo wrote:
> Hi,
> 
> 2018-12-28 12:30 Karsten Merker:
> > 
> > Debootstrap only uses packages from unstable so it cannot pull in
> > glibc 2.28 and stops with the error listed above.  One of
> > Debian's glibc maintainers (in CC) is working on the glibc 2.28
> > package to make it build cleanly for riscv64, so this problem
> > will hopefully be resolved in the near future.
> 
> Yes.  From what I remember, part of the reason why glibc 2.28 is not in
> the unstable archive "proper" is because of failures in glibc's test
> suite, which are really caused by the Linux kernel not cleaning up
> properly the state of FP registers between processes.  So they detect
> "real" problems :)
> 
> Aurélien submitted a patch for it earlier this month to linux-riscv
> mailing lists, which hasn't got any replies yet:
> 
>  http://lists.infradead.org/pipermail/linux-riscv/2018-December/002512.html
> 
> Not sure if he's waiting for a resolution to this problem.

Now that this problem has been understood and is identified as a kernel
issue and not a glibc one, I have whitelist those tests in our current
glibc git.

I am preparing a new upload, but I am first trying to debug a few other
issues in the testsuite which appear randomly. They do not happen often,
however given there are a few of them, the probability of failure is
still high. For debugging them, I am currently backporting gdb support
from upstream so that it's easier to debug the issue.

> The alternative is either retry building and see if in one occasion it
> builds fine and the tests pass due to "luck", and the new version gets
> added to the archive; or to disable these tests selectively (if
> feasible), or disable the test suite completely for riscv64.  In general
> we avoided to do the latter, since it helped to find other real problems
> also in the past.

I am planning to upload this new version in the next days. If it still
fails to build, we can indeed retry the build a few more times. That is
actually also useful to get statistics on the random issues.

Aurelien

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
aurelien at aurel32.net                 http://www.aurel32.net



More information about the lowrisc-dev mailing list