Re: [parisc-linux] -pie is broken on hppa

[email protected] a écrit :
> On Mon, Jun 11, 2007 at 11:16:59AM +0200, Aurelien Jarno wrote:
>> John David Anglin a écrit :
>>>> This is what I see when I do a gcc_update:
>>>> dave@mx3210:~/gnu/gcc-4.3/gcc$ contrib/gcc_update
>>>> Updating SVN tree
>>>> At revision 125275.
>>>> svn: Can't read directory '.': Partial results are valid but processing is incomplete
>>> I spent the day trying to debug why libjava fails to build on this system
>>> with the new glibc.  It seems that scandir is broken on 64-bit kernels.
>>> There seems to be an off by one error in copying dirent structs.  Either
>>> the define in <dirent.h> for _D_ALLOC_NAMLEN is wrong or the kernel
>>> sometimes returns a value for d_reclen that is sometimes too small.  The
>>> code in fs/readdir.c suggests that sys_parisc32.c needs fixing.
>>>
>>> The enclosed seems to fix both the libjava build and the svn problem.
>>> Kyle's branch uses ALIGN instead of ROUND_UP.
>> It looks like the glibc build problem may be due to the same bug. It
>> builds successfully on a 32-bit kernel, but fails to build on a 64-bit
>> kernel due to some sources files that are skipped during the build.
> 
> I tested the glibc build.
> Two things are worrying me.
> 
> 1- glibc builds with the linuxthread and not nptl. This lead to a missing
> symbol (__librt_multiple_threads) which let me think the activation of linuxthread
> maybe bad.
> The invokation is --enable-add-ons=libidn,ports linuxthreads. This part looks highly
> suspicious to me.

Building with linuxthreads is normal. The NPTL add-on currently breaks
the ABI for libpthread. This has to be fixed before we can really using it.

Note that I have the exact same build error with a 64-bit kernel, but
not with a 32-bit kernel. Also the glibc was building successfully on a
64-bit kernel until very recently. I guess the change from
linux-kernel-headers to linux-libc-dev (and with it a change from 2.6.18
to 2.6.21 headers) has triggered the kernel bug, just like it has
triggered the "df bug" (#427395).

> 2- when enabling the nptl add-on, the glibc builds ok but fails some tests. I don't
> really know if it is normal or not. How do I know there is regression or not ?

You have to compare the list of failed tests to the linuxthreads one. If
new tests are failing, you have to investigate why and see if this is a
regression or not.

Bye,
Aurelien

-- 
  .''`.  Aurelien Jarno	            | GPG: 1024D/F1BCDB73
 : :' :  Debian developer           | Electrical Engineer
 `. `'   [email protected]         | [email protected]
   `-    people.debian.org/~aurel32 | www.aurel32.net


-- 
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]