[Linux-ia64] gcc and glibc: big troubles

[Linux-ia64] gcc and glibc: big troubles

[Francis Galiegue]

I'm trying to upgrade my glibc and compiler to 001024 snapshot + RCS patches up
to 037 (for gcc) and glibc 2.2 plus patches (not CVS), but it fails on the
whole... Compiler used is 001024 snapshot but with backward compatibility ABI
patch applied.

First, glibc. I obtain this error:

-----
gcc -nostdlib -nostartfiles -o
/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/iconv/iconv_prog
-Wl,-dynamic-linker=/lib/ld-linux-ia64.so.2
/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/csu/crt1.o
/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/csu/crti.o `gcc
--print-file-name=crtbegin.o`
/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/iconv/iconv_prog.o
-Wl,-rpath-link=/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux:/usr/
src/RPM/BUILD/glibc-2.2/build-ia64-linux/math:/usr/src/RPM/BUILD/
glibc-2.2/build-ia64-linux/elf:/usr/src/RPM/BUILD/glibc-2.2/
build-ia64-linux/dlfcn:/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/
nss:/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/nis:/usr/src/RPM/
BUILD/glibc-2.2/build-ia64-linux/rt:/usr/src/RPM/BUILD/glibc-2.2/
build-ia64-linux/resolv:/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/
crypt:/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/linuxthreads
/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/libc.so.6.1
/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/libc_nonshared.a -lgcc `gcc
--print-file-name=crtend.o`
/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/csu/crtn.o

/usr/src/RPM/BUILD/glibc-2.2/build-ia64-linux/libc.so.6.1:
undefined reference to `__syscall_getpagesize'
collect2: ld returned 1 exit status
-----

Does CVS glibc cure this? If yes I guess I'll upgrade to the latest toolchain
snapshot too...

Now, trying to recompile the 001024 toolchain, but *without* the ABI patch, I
get the following:

-----
gcc -c -O2   -I../../../tk/unix -I../../../tk/unix/../generic
-I../../../tk/unix/../bitmaps -I/usr/src/RPM/BUILD/src/tcl/generic
-I/usr/X11R6/include -DHAVE_UNISTD_H=1 -DHAVE_LIMITS_H=1 -DSTDC_HEADERS=1
-DHAVE_SYS_TIME_H=1 -DTIME_WITH_SYS_TIME=1
../../../tk/unix/../generic/tkImgPhoto.c
../../../tk/unix/../generic/tkImgPhoto.c: In function `ImgPhotoCmd':
../../../tk/unix/../generic/tkImgPhoto.c:601: warning: passing arg 2 of
`Tcl_GetStringFromObj' from
incompatible pointer type
/tmp/ccalpirm.s: Assembler messages:
/tmp/ccalpirm.s:6579: Internal error, aborting at ../../gas/config/tc-ia64.c
line 5190 in errata_nop_necessary_p
Please report this bug.
-----

Culprit line:

-----
  if (insn_unit = IA64_UNIT_F)
    {
      for (i = 0; i < idesc->num_outputs; i++)
    if (idesc->operands[i] = IA64_OPND_P1
        || idesc->operands[i] = IA64_OPND_P2)
      {
        int regno = slot->opnd[i].X_add_number - REG_P;
        if (regno > 16)
          abort ();		<----- This one
        this_group->p_reg_set[regno] = 1;
      }
    }
-----

One thing I don't get is: why all this in the toolchain? Why make, why bison,
why tk, why tcl? Whats more there's no README at all on how to *avoid* having
them compiled. I tried --without-x, --disable-x but to no avail... Stock bison
for example works perfectly fine here, so does make. Any solution?

-- 
Francis Galiegue, fg@mandrakesoft.com
"Programming is a race between programmers, who try and make more and more
idiot-proof software, and universe, which produces more and more remarkable
idiots. Until now, universe leads the race"  -- R. Cook

Re: [Linux-ia64] gcc and glibc: big troubles

[Jes Sorensen]

>>>>> "Francis" = Francis Galiegue <fg@mandrakesoft.com> writes:

Francis> I'm trying to upgrade my glibc and compiler to 001024
Francis> snapshot + RCS patches up to 037 (for gcc) and glibc 2.2 plus
Francis> patches (not CVS), but it fails on the whole... Compiler used
Francis> is 001024 snapshot but with backward compatibility ABI patch
Francis> applied.

Francis> First, glibc. I obtain this error:

Please read the warning messages from the compiler when you compile
glibc and get errors like this. Higher up in the compiler output it
had warned you that it couldn't find the definition of
__NR_getpagesize(). It's because David took it out as we are switching
to AT_PAGESZ support instead. Until glibc gets fixed you can just
reenable it in include/asm/unistd.h.

Jes

Re: [Linux-ia64] gcc and glibc: big troubles

[H . J . Lu]

On Tue, Nov 21, 2000 at 05:00:44PM +0100, Jes Sorensen wrote:
> >>>>> "Francis" = Francis Galiegue <fg@mandrakesoft.com> writes:
> 
> Francis> I'm trying to upgrade my glibc and compiler to 001024
> Francis> snapshot + RCS patches up to 037 (for gcc) and glibc 2.2 plus
> Francis> patches (not CVS), but it fails on the whole... Compiler used
> Francis> is 001024 snapshot but with backward compatibility ABI patch
> Francis> applied.
> 
> Francis> First, glibc. I obtain this error:
> 
> Please read the warning messages from the compiler when you compile
> glibc and get errors like this. Higher up in the compiler output it
> had warned you that it couldn't find the definition of
> __NR_getpagesize(). It's because David took it out as we are switching
> to AT_PAGESZ support instead. Until glibc gets fixed you can just
> reenable it in include/asm/unistd.h.
> 

This is an untested patch. Let me know what you get.

Thanks.


H.J.
----
2000-11-21  H.J. Lu  <hjl@gnu.org>

	* sysdeps/unix/sysv/linux/ia64/getpagesize.c (__getpagesize):
	Don't call __syscall_getpagesize. Set to 8K by default.

	* sysdeps/unix/sysv/linux/ia64/syscalls.list (s_getpagesize):
	Removed.

Index: sysdeps/unix/sysv/linux/ia64/getpagesize.c
=================================RCS file: /work/cvs/gnu/glibc/sysdeps/unix/sysv/linux/ia64/getpagesize.c,v
retrieving revision 1.1.1.1
diff -u -p -r1.1.1.1 getpagesize.c
--- sysdeps/unix/sysv/linux/ia64/getpagesize.c	2000/06/25 16:02:54	1.1.1.1
+++ sysdeps/unix/sysv/linux/ia64/getpagesize.c	2000/11/21 16:48:07
@@ -27,19 +27,13 @@
    determine the page size to ensure proper alignment for calls such
    as mmap and friends.  --davidm 99/11/30 */
 
-/* If we are not a static program, this value is collected from the system
-   via the AT_PAGESZ auxiliary argument.  If we are a static program, we
-   use the getpagesize system call.  */
-
 extern size_t _dl_pagesize;
 
-extern size_t __syscall_getpagesize (void);
-
 int
 __getpagesize ()
 {
   if (_dl_pagesize = 0)
-    _dl_pagesize = INLINE_SYSCALL (getpagesize, 0);
+    _dl_pagesize = 8192;	/* use a reasonable default */
   return _dl_pagesize;
 }
 
Index: sysdeps/unix/sysv/linux/ia64/syscalls.list
=================================RCS file: /work/cvs/gnu/glibc/sysdeps/unix/sysv/linux/ia64/syscalls.list,v
retrieving revision 1.1.1.8
diff -u -p -r1.1.1.8 syscalls.list
--- sysdeps/unix/sysv/linux/ia64/syscalls.list	2000/09/16 06:41:37	1.1.1.8
+++ sysdeps/unix/sysv/linux/ia64/syscalls.list	2000/11/21 16:47:02
@@ -72,7 +72,6 @@ s_getdents64	getdents64 getdents64	i:ipi
 s_mknod		xmknod	mknod		i:sii	__syscall_mknod
 s_readv		readv	readv		i:ipi	__syscall_readv
 s_writev	writev	writev		i:ipi	__syscall_writev
-s_getpagesize	getpagesize getpagesize	0	__syscall_getpagesize
 s_poll		poll	poll		i:pii	__syscall_poll
 s_reboot	reboot	reboot		i:iii	__syscall_reboot
 s_ustat		ustat	ustat		i:ip	__syscall_ustat

Re: [Linux-ia64] gcc and glibc: big troubles

[Jim Wilson]

>/tmp/ccalpirm.s:6579: Internal error, aborting at ../../gas/config/tc-ia64.c
>line 5190 in errata_nop_necessary_p
>        if (regno > 16)
>          abort ();		<----- This one

This is a bug in the hardware errata workaround patch.  This should be
	if (regno >= 64)
	  abort ();
I found this problem in pre-release testing, so it is fixed in the 001117
release, but I didn't release a separate patch for this bug fix.

>One thing I don't get is: why all this in the toolchain? Why make, why bison,
>why tk, why tcl? Whats more there's no README at all on how to *avoid* having
>them compiled. I tried --without-x, --disable-x but to no avail... Stock bison
>for example works perfectly fine here, so does make. Any solution?

This is the "Cygnus" way of making releases.  This is necessary for "Cygnus"
to make sure that our customers can rebuild our toolchains.  Otherwise,
they may not have the necessary tools (e.g. GNU make), or they may have
incompatible versions of them (tcl/tk).

This style of release is not appropriate for GNU/Linux distributors, but
unfortunately, we are stuck with it for now.  I'd like to get people moved
to FSF style releases as soon as possible, but I couldn't easily do that until
after most of the development work was done.  I expect the next release will be
from the FSF gcc/binutils/gdb sources, so this should not be a problem with
the next release.

tcl/tk/tix/itcl are there for the gdb gui.  This requires the specific versions
of tcl/tk/etc in the toolchain, otherwise you will run into versioning
problems.  The reasons for the rest of the stuff is probably obvious, they
all come standard with linux, but not with SunOS/Solaris/AIX/Irix/NT/etc.

You can build specific tools, like gcc, by using
    make all-gcc
There are makefile dependencies which will force everything gcc requires
(opcodes/bfd/binutils/gas/ld) to also be built.  You can build specific
target libraries, like libstdc++, by using
    make all-target-libstdc++
Making all-gcc, all-gdb, and all-target-libstdc++, will probably build
everything you want.

You can install specific tools, like gcc, by using
   make install-gcc
There are no dependencies here, so you need to separately install everything
you want, e.g. bfd, binutils, gas, ld, gdb, gcc, target-libstdc++, and maybe
some other stuff.

You can disable building the gdb gui by using --disable-gdbtk

You can also disable building stuff by renaming/deleting the directories.
This will work for all of the optional stuff like flex/bison/make/patch/etc
that comes standard with Linux and which are unlikely to suffer versioning
problems.

All of this stuff I just described also works exactly the same way in FSF
gcc/binutils/gdb releases.  We just bundle everything together instead of
shipping everything separately.

There is documentation explaining all of this stuff, but it is maintained
by the documentation group not the engineers, and is usually distributed on
paper.  A "Cygnus" customer would get it, but it is hard to distribute to
everyone trying to use my Trillian releases.

I like to point out again that my releases are primarily for the Trillian
group, to enable kernel, glibc, and Linux distribution work.  It was never
the intent to recommend them to everyone.  I'd rather recommend the compilers
that come with the Linux distributions, but unfortunately, all IA-64 GNU/Linux
distributions are still using the old February toolchain, so as a stop gap
measure I'm making my toolchains available to everyone.  Since you are
Linux distributor though, you'd probably have to build my toolchain anyways.

Jim

Re: [Linux-ia64] gcc and glibc: big troubles

[Francis Galiegue]

On Tue, 21 Nov 2000, Jim Wilson wrote:

Jim,

Thanks very much for your complete answer. I'm afraid I'm not as good as you
guys, but I'm doing my best to follow :)

> 
> >/tmp/ccalpirm.s:6579: Internal error, aborting at ../../gas/config/tc-ia64.c
> >line 5190 in errata_nop_necessary_p
> >        if (regno > 16)
> >          abort ();		<----- This one
> 
> This is a bug in the hardware errata workaround patch.  This should be
> 	if (regno >= 64)
> 	  abort ();
> I found this problem in pre-release testing, so it is fixed in the 001117
> release, but I didn't release a separate patch for this bug fix.
> 

OK, I made a patch over the 001124 release... Compiling with it right now.

> 
> You can build specific tools, like gcc, by using
>     make all-gcc
> There are makefile dependencies which will force everything gcc requires
> (opcodes/bfd/binutils/gas/ld) to also be built.  You can build specific
> target libraries, like libstdc++, by using
>     make all-target-libstdc++
> Making all-gcc, all-gdb, and all-target-libstdc++, will probably build
> everything you want.
> 
> You can install specific tools, like gcc, by using
>    make install-gcc
> There are no dependencies here, so you need to separately install everything
> you want, e.g. bfd, binutils, gas, ld, gdb, gcc, target-libstdc++, and maybe
> some other stuff.
> 

Thanks, rebuilding with these right now.

> You can disable building the gdb gui by using --disable-gdbtk
> 

Already in my spec file.

> 
> I like to point out again that my releases are primarily for the Trillian
> group, to enable kernel, glibc, and Linux distribution work.  It was never
> the intent to recommend them to everyone.  I'd rather recommend the compilers
> that come with the Linux distributions, but unfortunately, all IA-64 GNU/Linux
> distributions are still using the old February toolchain, so as a stop gap
> measure I'm making my toolchains available to everyone.  Since you are
> Linux distributor though, you'd probably have to build my toolchain anyways.
> 

Yep, very useful for building from a chroot... Thanks for your time again.

-- 
Francis Galiegue, fg@mandrakesoft.com
"Programming is a race between programmers, who try and make more and more
idiot-proof software, and universe, which produces more and more remarkable
idiots. Until now, universe leads the race"  -- R. Cook