[Bug c/24279] New: SEGV at reload.c:2400 with -O2

[Bug c/24279] New: SEGV at reload.c:2400 with -O2

[shap at eros-os dot org]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 2468 bytes --]

I am cross-building a compiler, hosted on FC4 (i686-pc-linux-gnu) and targeting
m68k-elf. The following bug is reproducible in gcc-4.0.1 and gcc-4.0.2. The
offending line number is in gcc-4.0.2. The bug causes build of newlib to fail.

Here is how I am building the cross tool chain. I am aware that this is not
ideal, but it is simpler than the real build for my actual target
(m68k-unknown-coyotos). The failure is reproducible using the instructions
below.

# Unpack binutils-2.16.1
# make an empty directory, configure as
{BINUTILS_SRC}/configure --prefix=/usr/local/m68k-elf --target m68k-elf

# Build binutils as follows:
make tooldir=/usr/local/m68k-elf all
make install

#Unpack gcc-core-4.0.2 from FSF
# configure with:
{GCC_SRC}/configure --prefix=/usr/local/m68k-elf --without-headers
--with-newlib --disable-libmudflap --enable-languages=c --target m68k-elf

# build with:
PATH=/usr/local/m68k-elf/bin:${PATH} make gcc_tooldir=/usr/local/m68k-elf/bin
all-gcc
PATH=/usr/local/m68k-elf/bin:${PATH} make gcc_tooldir=/usr/local/m68k-elf/bin
install-gcc


This build proceeds successfully. Following the build, I attempt to execute
(vfprintf.i attached):

/usr/local/m68k-elf/libexec/gcc/m68k-elf/4.0.2/cc1  -fpreprocessed vfprintf.i
-quiet -dumpbase vfprintf.c -m68000 -m68000 -auxbase vfprintf -g -O2 -version
-fno-builtin -fshort-enums -o vfprintf.s
GNU C version 4.0.2 (m68k-elf)
        compiled by GNU C version 4.0.1 20050727 (Red Hat 4.0.1-5).
GGC heuristics: --param ggc-min-expand=99 --param ggc-min-heapsize=129233
../../../../../../newlib-1.13.0/newlib/libc/stdio/vfprintf.c: In function
‘_vfprintf_r’:
../../../../../../newlib-1.13.0/newlib/libc/stdio/vfprintf.c:1220: internal
compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://gcc.gnu.org/bugs.html> for instructions.


Compilation succeeds if '-O2' is replaced with '-O1' or '-O'. Turning off the
various -f options has no impact.

I am more than happy to test proposed patches if helpful.


-- 
           Summary: SEGV at reload.c:2400 with -O2
           Product: gcc
           Version: 4.0.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: shap at eros-os dot org
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: m68k-elf


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug c/24279] SEGV at reload.c:2400 with -O2

[shap at eros-os dot org]

------- Comment #1 from shap at eros-os dot org  2005-10-08 21:18 -------
Created an attachment (id=9942)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=9942&action=view)
preprocessed vfprintf.c from newlib for this target.

This is the input file that I am trying to compile. I have not attempted to
chop it down.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug c/24279] SEGV at reload.c:2400 with -O2

[shap at eros-os dot org]

------- Comment #2 from shap at eros-os dot org  2005-10-08 21:20 -------
Setting aside whether this particular piece of code is broken (it may be, but
it is a pretty widely used part of newlib), it seems unlikely that a SIGSEGV in
the optimizer would be the preferred diagnostic. ;-)

Jokes aside, I'm set up to rebuild my target quickly, so if there is a patch to
test or a further experiment to run please do not hesitate to ask.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug middle-end/24279] SEGV at reload.c:2400 with -O2

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2005-10-08 21:24 -------
I think this might be a duplicate of PR 23043 which is fixed for 4.0.3, it
missed 4.0.2 by a mistake.  Can you try a cvs version of the 4.0 branch to see
if this is fixed? Or try the patch which is referenced in that PR?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug middle-end/24279] SEGV at reload.c:2400 with -O2

[shap at eros-os dot org]

------- Comment #4 from shap at eros-os dot org  2005-10-08 21:58 -------
Testing the patch now. I'll report back shortly.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug middle-end/24279] SEGV at reload.c:2400 with -O2

[shap at eros-os dot org]

------- Comment #5 from shap at eros-os dot org  2005-10-08 22:19 -------
The patch from PR 23043 does NOT resolve this. Same error, reported at the same
place in reload.c.

Given the number of other mode-related bugs against m68k, I was suspicious of
that, but when I re-run under gdb, the value of val.start at the time of the
fault is improbably high (1708), and then I get

(gdb) p hard_regno_nregs[val.start]
Cannot access memory at address 0x831b0d0

Not sure if this helps suggest anything.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug middle-end/24279] SEGV at reload.c:2400 with -O2

[shap at eros-os dot org]

------- Comment #6 from shap at eros-os dot org  2005-11-05 19:44 -------
I know you folks have many other things to do, but any further ideas on this
one?  If not, what can I do to help get the test case confirmed and into the
regression suite?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug middle-end/24279] SEGV at reload.c:2400 with -O2

[kazu at gcc dot gnu dot org]

------- Comment #7 from kazu at gcc dot gnu dot org  2005-11-19 21:13 -------
Jonathan,

The testcase seems to work with GCC 4.1.

As far as getting the testcase into the testsuite, you need to post a patch
first.
The best thing to do at this point is to reduce the testcase so that it will be
easier to analyze what's going on.


-- 

kazu at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kazu at gcc dot gnu dot org
      Known to work|                            |4.1.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279

[Bug middle-end/24279] SEGV at reload.c:2400 with -O2

[kazu at gcc dot gnu dot org]

------- Comment #8 from kazu at gcc dot gnu dot org  2006-05-08 16:32 -------
Works as of 4.1 according to #7.


-- 

kazu at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=24279