Ports state for 2.16 regarding global changes

Ports state for 2.16 regarding global changes

[Joseph S. Myers]

Here is my understanding of the current status of ports with regard to
being updated for global changes in libc:

* arm, ia64, linux-generic, mips, powerpc-nofpu, tile: up to date.
  However, ia64 libm-test-ulps may need regenerating / updating.

* m68k: mostly up to date, but may need updates to its ccos / csin /
  ccosh / csinh implementations along the lines of my 2012-05-19 libc
  changes to avoid spurious overflows (bug 14123), and both m68k
  libm-test-ulps files may need regenerating / updating.

* alpha: out of date in the following ways:

  (a) There are __GNUC_PREREQ conditionals in various non-installed
  files that should be removed because the versions in question are
  older than the minimum 4.3 version accepted for configuring GCC:
  sysdeps/alpha/fpu/cfloat-compat.h sysdeps/alpha/fpu/s_fabs.c
  sysdeps/alpha/fpu/s_fabsf.c sysdeps/unix/sysv/linux/alpha/sysconf.c.

  (b) bits/mathdef.h should be updated in line with my 2012-01-31 libc
  change so that float_t is always float rather than depending on
  __GNUC__.

  (c) sysdeps/alpha/soft-fp/sfp-machine.h is missing a definition of
  FP_TRAPPING_EXCEPTIONS.

  (d) sysdeps/unix/sysv/linux/alpha/bits/resource.h is missing
  RLIMIT_RTTIME.

  (e) sysdeps/unix/sysv/linux/alpha/bits/resource.h has a possible
  trailing comma in an enum outside __USE_GNU where the generic code
  has it inside __USE_GNU.

  (f) sysdeps/unix/sysv/linux/alpha/bits/resource.h is missing the
  prlimit and prlimit64 declarations.

  (g) sysdeps/unix/sysv/linux/alpha/bits/shm.h is missing SHM_EXEC.

  (h) sysdeps/unix/sysv/linux/alpha/bits/stat.h has a conditional "#if
  defined __USE_ATFILE || defined __USE_GNU" where other versions use
  "#ifdef __USE_ATFILE".

* hppa: out of date in the following ways:

  (a) There is a stray file
  sysdeps/unix/sysv/linux/hppa/nptl/pt-initfini.c, which will be unused.

  (b) libm-test-ulps needs to be regenerated / updated.

  (c) There is no ____longjmp_chk implementation.

  (d) sysdeps/unix/sysv/linux/hppa/nptl/pthread_once.c has an alias
  __pthread_once_internal but this has been replaced by hidden_def for
  other architectures.

  (e) sysdeps/unix/sysv/linux/hppa/syscalls.list has a
  __connect_internal alias that has been removed for other
  architectures.

  (f) sysdeps/unix/sysv/linux/hppa/nptl/pthread.h uses
  __builtin_expect but the generic version now uses __glibc_unlikely
  instead.

  (g) There are no ABI test baselines for hppa.

  (h) hppa still has conditionals on or definitions of obsolete macros
  USE_TLS and USE___THREAD.

  (i) sysdeps/unix/sysv/linux/hppa/bits/errno.h is missing EHWPOISON
  (and any associated errlist-compat updates needed).  A patch was
  submitted in
  <http://sourceware.org/ml/libc-ports/2012-02/msg00061.html> but
  apparently not reviewed.

  (j) sysdeps/unix/sysv/linux/hppa/bits/fcntl.h is missing O_PATH.

  (k) sysdeps/unix/sysv/linux/hppa/bits/mman.h is missing MAP_STACK
  and MAP_HUGETLB.

  (l) sysdeps/unix/sysv/linux/hppa/bits/mman.h is missing
  MADV_HUGEPAGE, MADV_NOHUGEPAGE, MADV_DONTDUMP, MADV_DODUMP and
  MADV_HWPOISON.

  (m) sysdeps/unix/sysv/linux/hppa/bits/shm.h is missing SHM_EXEC.

  (n) sysdeps/unix/sysv/linux/hppa/bits/sigaction.h is missing the
  change to define SA_RESTART, SA_NODEFER and SA_RESETHAND if
  __USE_XOPEN2K8.

  (o) sysdeps/unix/sysv/linux/hppa/bits/socket.h still exists but
  socket_type.h should now be provided instead.

  (p) sysdeps/unix/sysv/linux/hppa/nptl/pthread.h defines
  __need_timespec before including <time.h> but the generic version no
  longer does.

  (q) sysdeps/unix/sysv/linux/hppa/sys/epoll.h still exists but
  bits/epoll.h should now be provided instead.

  (r) sysdeps/unix/sysv/linux/hppa/sys/eventfd.h still exists but
  bits/eventfd.h should now be provided instead.

  (s) sysdeps/unix/sysv/linux/hppa/sys/inotify.h still exists but
  bits/inotify.h should now be provided instead.

  (t) sysdeps/unix/sysv/linux/hppa/sys/signalfd.h still exists but
  bits/signalfd.h should now be provided instead.

  (u) sysdeps/unix/sysv/linux/hppa/sys/timerfd.h still exists but
  bits/timerfd.h should now be provided instead.

  (v) sysdeps/unix/sysv/linux/hppa/syscalls.list is missing prlimit64
  and fanotify_mark which means it is likely those functions are
  missing on hppa (and when added, will need appropriate Versions
  additions to avoid adding them to an old version).

  (w) An entry in libc's nptl/shlib-versions for hppa was removed on
  2012-01-07 and a corresponding entry needs adding in ports.

  (x) An entry in libc's sysdeps/unix/sysv/linux/configure.in for hppa
  was removed on 2012-01-07 and a ports update may be needed.  That
  entry may actually have been dead because of the arch_minimum_kernel
  setting in sysdeps/unix/sysv/linux/hppa/nptl/configure.in - but in
  that case, there's the question of whether the
  libc_cv_gcc_unwind_find_fde=yes that was in libc's file for hppa
  should go in the ports version (if that was a correct description of
  old libc binaries on hppa).

  (y) sysdeps/unix/sysv/linux/hppa/kernel-features.h still defines
  __ASSUME_32BITUIDS and __ASSUME_TRUNCATE64_SYSCALL (no longer used).

* am33: generally massively out of date and I have not tried to
  identify specific changes needed.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: Ports state for 2.16 regarding global changes

[Mark Salter]

On Fri, 2012-06-15 at 15:52 +0000, Joseph S. Myers wrote:
> * am33: generally massively out of date and I have not tried to
>   identify specific changes needed.

I have been working on getting this updated. ATM, I'm blocked waiting on
a fix for a bug in the gcc patch adding am33 tls support. I expect to be
past this sometime next week so I can actually start testing things.

Would it be helpful to go ahead and post the glibc patches now or wait
until I get them tested?

--Mark

Re: Ports state for 2.16 regarding global changes

[Joseph S. Myers]

On Fri, 15 Jun 2012, Mark Salter wrote:

> On Fri, 2012-06-15 at 15:52 +0000, Joseph S. Myers wrote:
> > * am33: generally massively out of date and I have not tried to
> >   identify specific changes needed.
> 
> I have been working on getting this updated. ATM, I'm blocked waiting on
> a fix for a bug in the gcc patch adding am33 tls support. I expect to be
> past this sometime next week so I can actually start testing things.
> 
> Would it be helpful to go ahead and post the glibc patches now or wait
> until I get them tested?

I think it makes sense to post them now.  I guess you'll have a small 
patch to libc - adding new relocations to elf.h - which should be sent to 
libc-alpha and should be safe even during the freeze, and the rest should 
be for ports.  If the patches pass review then I think we should give you 
write access and make you am33 maintainer so you can apply further am33 
patches without prior review - given the present state of am33 support 
(i.e. not substantively maintained since 2004), the changes are hardly 
going to make things any worse.

(Updates to libc's README and NEWS files of course wait until the port is 
actually working.)

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: Ports state for 2.16 regarding global changes

[Richard Henderson]

On 2012-06-15 08:52, Joseph S. Myers wrote:
>   (c) sysdeps/alpha/soft-fp/sfp-machine.h is missing a definition of
>   FP_TRAPPING_EXCEPTIONS.

What's this one supposed to do?  There's no commentary anywhere.


r~

Re: Ports state for 2.16 regarding global changes

[Joseph S. Myers]

On Fri, 15 Jun 2012, Richard Henderson wrote:

> On 2012-06-15 08:52, Joseph S. Myers wrote:
> >   (c) sysdeps/alpha/soft-fp/sfp-machine.h is missing a definition of
> >   FP_TRAPPING_EXCEPTIONS.
> 
> What's this one supposed to do?  There's no commentary anywhere.

It expands to a bitmask of those exceptions (FP_EX_* values) for which 
trapping (sending a signal) is enabled.  This is only relevant for cases 
of exact underflow (where a signal is sent if trapping on underflow is 
enabled, but otherwise the underflow flag is not raised).

Typically a masked version of the floating-point control register, shifted 
to move the trap-enabled bits into the positions of the exception-raised 
bits, is right.

-- 
Joseph S. Myers
joseph@codesourcery.com