Re: tilepro glibc problems

[Chris Metcalf <cmetcalf at mellanox dot com>]

On 11/4/2016 4:05 PM, Joseph Myers wrote:
> Compile testing of glibc for tilepro (GCC 5, binutils master) shows up two
> problems:
>
> * C++ linking doesn't work (which causes C++ tests in glibc to be
> disabled):
>
> /scratch/jmyers/glibc/many5/install/compilers/tilepro-linux-gnu/lib/gcc/tilepro-glibc-linux-gnu/5.4.1/../../../../tilepro-glibc-linux-gnu/bin/ld:
> a.out: hidden symbol `__atomic_fetch_add_4' in
> /scratch/jmyers/glibc/many5/install/compilers/tilepro-linux-gnu/lib/gcc/tilepro-glibc-linux-gnu/5.4.1/libgcc.a(atomic.o)
> is referenced by DSO
> /scratch/jmyers/glibc/many5/install/compilers/tilepro-linux-gnu/lib/gcc/tilepro-glibc-linux-gnu/5.4.1/../../../../tilepro-glibc-linux-gnu/bin/ld:
> final link failed: Bad value
> collect2: error: ld returned 1 exit status
>
> (This seems to me like your GCC port should be using t-slibgcc-libgcc in
> libgcc/config.host, but I haven't tested that.)

Thanks.  I'll forward that on to our compiler guys.

> * sys/dataplane.h includes a header asm/dataplane.h and thereby fails the
> check-installed-headers test because no such header exists in the Linux
> kernel.  How is this meant to work?  I imagine this would fail for tilegx
> as well if that build didn't fail at an earlier stage.

This header was submitted basically as part of the original glibc patch
series, overlooking the fact that kernel support was not yet upstreamed.

It was intended to support the task isolation patch series in the kernel [1],
but the sys/dataplane.h header targets what you might call the "v0" of that
patch series, namely the version that Tilera shipped customers prior to
starting to upstream it.  (At that time it was called "dataplane" support.)

The current API proposal is handled entirely out of linux/prctl.h, so this
header will no longer be necessary; I'll just go ahead and remove it.

--
Chris Metcalf, Mellanox Technologies
http://www.mellanox.com