[Bug target/63691] New: GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[Bug target/63691] New: GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[aaro.koskinen at iki dot fi]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

            Bug ID: 63691
           Summary: GCC 4.9.x fails to build GLIBC 2.20 on HPPA
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: aaro.koskinen at iki dot fi

GCC 4.8.3 builds GLIBC 2.20 (w/Gentoo patches) fine, but with GCC 4.9 it fails:

hppa-linux-gnu-gcc   -nostdlib -nostartfiles -r -o
/home/aaro/los/work/parisc/glibc-2.20-build/default/elf/librtld.map.o '-Wl,-('
/home/aaro/los/work/parisc/glibc-2.20-build/default/elf/dl-allobjs.os
/home/aaro/los/work/parisc/glibc-2.20-build/default/libc_pic.a -lgcc '-Wl,-)'
-Wl,-Map,/home/aaro/los/work/parisc/glibc-2.20-build/default/elf/librtld.mapT
/home/aaro/los/work/parisc/glibc-2.20-build/default/libc_pic.a(dl-addr.os): In
function `_dl_addr_inside_object':
/home/aaro/los/work/shared/glibc-2.20/elf/dl-addr.c:152: multiple definition of
`_dl_addr_inside_object'
/home/aaro/los/work/parisc/glibc-2.20-build/default/elf/dl-allobjs.os:(.text+0x8b8):
first defined here
/home/aaro/los/work/parisc/glibc-2.20-build/default/libc_pic.a(init-first.os):(.data+0x0):
multiple definition of `__libc_multiple_libcs'
/home/aaro/los/work/parisc/glibc-2.20-build/default/elf/dl-allobjs.os:(.bss+0x7c):
first defined here
/home/aaro/los/work/parisc/glibc-2.20-build/default/libc_pic.a(_itoa.os): In
function `_itoa':
/home/aaro/los/work/shared/glibc-2.20/stdio-common/_itoa.c:199: multiple
definition of `_itoa'
/home/aaro/los/work/parisc/glibc-2.20-build/default/elf/dl-allobjs.os:(.text+0x5b8):
first defined here
collect2: error: ld returned 1 exit status

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[aaro.koskinen at iki dot fi]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

Aaro Koskinen <aaro.koskinen at iki dot fi> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |carlos at systemhalted dot org,
                   |                            |dave.anglin at bell dot net,
                   |                            |law at redhat dot com

--- Comment #1 from Aaro Koskinen <aaro.koskinen at iki dot fi> ---
Bisected on gcc git gcc-4_9-branch to:

db242b6d0eba96a468a2cea2c0e175aeb210644a is the first bad commit
commit db242b6d0eba96a468a2cea2c0e175aeb210644a
Author: law <law@138bc75d-0d04-0410-961f-82ee72b054a4>
Date:   Tue Nov 5 19:47:44 2013 +0000

        * Makefile.in (OBJS): Add gimple-ssa-isolate-paths.o

Unfortunately this does not revert cleanly so I cannot verify the result of
bisection by reverting the commit.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #3 from dave.anglin at bell dot net ---
On 30-Oct-14, at 5:19 PM, aaro.koskinen at iki dot fi wrote:

> Bisected on gcc git gcc-4_9-branch to:

Would you please attach .i and .s files generated from dl-addr.c?

Dave
--
John David Anglin    dave.anglin@bell.net

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
this was a bug in glibc iirc. it showed up on aarch64 too.  there is two ways
of solving it, backport the patch to glibc or define trap pattern in the pa
backend.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
See https://sourceware.org/ml/libc-alpha/2013-11/msg00291.html .

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[aaro.koskinen at iki dot fi]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #6 from Aaro Koskinen <aaro.koskinen at iki dot fi> ---
(In reply to Andrew Pinski from comment #5)
> See https://sourceware.org/ml/libc-alpha/2013-11/msg00291.html .

This patch is already included in GLIBC 2.20...

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[dave.anglin at bell dot net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #7 from dave.anglin at bell dot net ---
On 10/31/2014 2:39 PM, aaro.koskinen at iki dot fi wrote:
> This patch is already included in GLIBC 2.20...
I can add support for __builtin_trap() but we need preprocessed source
and full compile command.  You can add "-save-temps -v" to glibc command
to generate these files.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[aaro.koskinen at iki dot fi]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #8 from Aaro Koskinen <aaro.koskinen at iki dot fi> ---
(In reply to dave.anglin from comment #7)
> On 10/31/2014 2:39 PM, aaro.koskinen at iki dot fi wrote:
> > This patch is already included in GLIBC 2.20...
> I can add support for __builtin_trap() but we need preprocessed source
> and full compile command.  You can add "-save-temps -v" to glibc command
> to generate these files.

I was able to isolate the problem to GLIBC elf/dl-conflict.c and specifically
hppa specific elf_machine_rela function called from there. The problem occurs
at link time, so crafting a standalone preprocessed source file triggering the
error might be difficult, but I'll see what can I do.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[schwab@linux-m68k.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #9 from Andreas Schwab <schwab@linux-m68k.org> ---
You don't need a link test, you only need to check whether there is a reference
to abort.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[aaro.koskinen at iki dot fi]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #10 from Aaro Koskinen <aaro.koskinen at iki dot fi> ---
(In reply to Andreas Schwab from comment #9)
> You don't need a link test, you only need to check whether there is a
> reference to abort.

A simple test:

$ cat pr63691.c
void foo (int a, char *p)
{
        if (a == 0)
                return;
        *p = 1;
}
int main (int argc, char **argv)
{
        foo(argc, 0);
        return 0;
}
$ hppa-linux-gnu-gcc -nostdlib -O2 pr63691.c
/home/aaro/los/work/parisc/toolchain/lib/gcc/hppa-linux-gnu/4.9.2/../../../../hppa-linux-gnu/bin/ld:
warning: cannot find entry symbol _start; defaulting to 0000000000010054
/tmp/ccWAW31H.o: In function `main':
(.text.startup+0x14): undefined reference to `abort'
collect2: error: ld returned 1 exit status

abort() reference is gone if you compile GCC with the patch
Dave posted here:

https://www.sourceware.org/ml/libc-alpha/2014-10/msg00823.html

BTW, I think this optimization also breaks PALO (PA-RISC bootloader). There the
program actually wants to access page zero.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
   Last reconfirmed|                            |2014-11-01
                 CC|                            |danglin at gcc dot gnu.org
         Resolution|INVALID                     |---
     Ever confirmed|0                           |1

--- Comment #11 from John David Anglin <danglin at gcc dot gnu.org> ---
There is a reference to abort and abort is part of stdlib.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[aaro.koskinen at iki dot fi]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

--- Comment #13 from Aaro Koskinen <aaro.koskinen at iki dot fi> ---
Created attachment 33884
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33884&action=edit
Preprocessed dl-conflict

Here's the preprocessed dl-conflict.c. In elf_machine_rela(), the sym_map gets
always assigned to 0, and in the following switch cases it's dereferenced
without any checks. Those are where the compiler will add abort refs.

When compiled with -fno-delete-null-pointer-checks GCC won't do that, but it's
not really related to any null pointer checks, instead the switch has also
other meanings, i.e. it's also telling the compiler that NULL deref is not
erroneous so it cannot make program fail on those. Or at least that's how I
understand it.

[Bug target/63691] GCC 4.9.x fails to build GLIBC 2.20 on HPPA

[danglin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63691

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #15 from John David Anglin <danglin at gcc dot gnu.org> ---
Fixed.