[Bug target/45807] New: Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[Bug target/45807] New: Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[amodra at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

           Summary: Lying eh_frame r2 save info causes crashes with static
                    libgcc_eh and libstdc++
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: amodra@gmail.com


Created attachment 21893
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=21893
testcase from Allan Pratt <akpratt@netflix.com>

In the eh_frame description for _Unwind_Resume_or_Rethrow, I see

000012a0 0000007c 000012a4 FDE cie=00000000 pc=100712c0..100714f8
  DW_CFA_advance_loc: 4 to 100712c4
  DW_CFA_def_cfa_offset: 3216
  DW_CFA_advance_loc: 164 to 10071368
  DW_CFA_offset_extended_sf: r2 at cfa+40
[snip]

but, r2 is not saved by this function, at least, not before
_Unwind_RaiseException is called.  So the lie that r2 is saved at
cfa+40 once we are past 10071368 can cause the unwinder to return a bad
value for r2 when unwinding through _Unwind_Resume_or_Rethrow from 
_Unwind_RaiseException.  r2 won't be saved if _Unwind_Resume_of_Rethrow is
called statically from the same toc group, but this bug is often covered by the
fact that a previous call into exception handling functions has written the
correct r2 value to the stack.

00000000100712c0 <._Unwind_Resume_or_Rethrow>:
    100712c0:   f8 21 f3 71     stdu    r1,-3216(r1)
    100712c4:   7d 80 00 26     mfcr    r12
    100712c8:   7c 08 02 a6     mflr    r0
    100712cc:   d9 c1 0c 00     stfd    f14,3072(r1)
    100712d0:   d9 e1 0c 08     stfd    f15,3080(r1)
    100712d4:   f8 01 0c a0     std     r0,3232(r1)
    100712d8:   da 01 0c 10     stfd    f16,3088(r1)
    100712dc:   da 21 0c 18     stfd    f17,3096(r1)
    100712e0:   da 41 0c 20     stfd    f18,3104(r1)
    100712e4:   da 61 0c 28     stfd    f19,3112(r1)
    100712e8:   da 81 0c 30     stfd    f20,3120(r1)
    100712ec:   da a1 0c 38     stfd    f21,3128(r1)
    100712f0:   da c1 0c 40     stfd    f22,3136(r1)
    100712f4:   da e1 0c 48     stfd    f23,3144(r1)
    100712f8:   db 01 0c 50     stfd    f24,3152(r1)
    100712fc:   db 21 0c 58     stfd    f25,3160(r1)
    10071300:   db 41 0c 60     stfd    f26,3168(r1)
    10071304:   db 61 0c 68     stfd    f27,3176(r1)
    10071308:   db 81 0c 70     stfd    f28,3184(r1)
    1007130c:   db a1 0c 78     stfd    f29,3192(r1)
    10071310:   db c1 0c 80     stfd    f30,3200(r1)
    10071314:   db e1 0c 88     stfd    f31,3208(r1)
    10071318:   f9 c1 0b 70     std     r14,2928(r1)
    1007131c:   f9 e1 0b 78     std     r15,2936(r1)
    10071320:   fa 01 0b 80     std     r16,2944(r1)
    10071324:   fa 21 0b 88     std     r17,2952(r1)
    10071328:   fa 41 0b 90     std     r18,2960(r1)
    1007132c:   fa 61 0b 98     std     r19,2968(r1)
    10071330:   fa 81 0b a0     std     r20,2976(r1)
    10071334:   fa a1 0b a8     std     r21,2984(r1)
    10071338:   fa c1 0b b0     std     r22,2992(r1)
    1007133c:   fa e1 0b b8     std     r23,3000(r1)
    10071340:   fb 01 0b c0     std     r24,3008(r1)
    10071344:   fb 21 0b c8     std     r25,3016(r1)
    10071348:   fb 41 0b d0     std     r26,3024(r1)
    1007134c:   fb 61 0b d8     std     r27,3032(r1)
    10071350:   fb 81 0b e0     std     r28,3040(r1)
    10071354:   fb a1 0b e8     std     r29,3048(r1)
    10071358:   fb c1 0b f0     std     r30,3056(r1)
    1007135c:   fb e1 0b f8     std     r31,3064(r1)
    10071360:   e8 03 00 10     ld      r0,16(r3)
    10071364:   7c 7f 1b 78     mr      r31,r3
    10071368:   2f a0 00 00     cmpdi   cr7,r0,0
    1007136c:   f8 61 0b 50     std     r3,2896(r1)
    10071370:   f8 81 0b 58     std     r4,2904(r1)
    10071374:   f8 a1 0b 60     std     r5,2912(r1)
    10071378:   f8 c1 0b 68     std     r6,2920(r1)
    1007137c:   91 81 0c 98     stw     r12,3224(r1)
    10071380:   40 9e 00 e0     bne-    cr7,10071460
<._Unwind_Resume_or_Rethrow+0x1a0>
    10071384:   4b ff f8 0d     bl      10070b90 <._Unwind_RaiseException>

[Bug target/45807] Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[amodra at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

Alan Modra <amodra at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2010.09.27 10:27:49
               date|                            |
     Ever Confirmed|0                           |1

[Bug target/45807] Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[amodra at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

--- Comment #2 from Alan Modra <amodra at gcc dot gnu.org> 2010-09-28 15:25:08 UTC ---
Author: amodra
Date: Tue Sep 28 15:25:03 2010
New Revision: 164685

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=164685
Log:
    PR target/45807
    * config/rs6000/aix.h (SETUP_FRAME_ADDRESSES): Delete.
    * config/rs6000/linux64.h (SETUP_FRAME_ADDRESSES): Delete.
    * config/rs6000/rs6000-protos.h (rs6000_aix_emit_builtin_unwind_init):
    Delete.
    * config/rs6000/rs6000.c (rs6000_aix_emit_builtin_unwind_init): Delete.
    (rs6000_emit_prologue): Don't just create frame save info for r2,
    actually save r2.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/rs6000/aix.h
    trunk/gcc/config/rs6000/linux64.h
    trunk/gcc/config/rs6000/rs6000-protos.h
    trunk/gcc/config/rs6000/rs6000.c

[Bug target/45807] Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[meissner at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

Michael Meissner <meissner at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |meissner at gcc dot gnu.org

--- Comment #3 from Michael Meissner <meissner at gcc dot gnu.org> 2010-09-30 19:04:09 UTC ---
This breaks builds where the default is 64-bit, i.e. --with-cpu=default64

With --with-cpu=default64, it fails in building libgcc.a:
/home/meissner/fsf-install-ppc64/binutils-current/bin/ranlib libgcc.a
/home/meissner/fsf-build-ppc64/trunk/./gcc/xgcc
-B/home/meissner/fsf-build-ppc64/trunk/./gcc/
-B/home/meissner/fsf-install-ppc64/trunk/powerpc64-linux/bin/
-B/home/meissner/fsf-install-ppc64/trunk/powerpc64-linux/lib/ -isystem
/home/meissner/fsf-install-ppc64/trunk/powerpc64-linux/include -isystem
/home/meissner/fsf-install-ppc64/trunk/powerpc64-linux/sys-include    -g -O2
-O2  -g -O2 -DIN_GCC   -W -Wall -Wwrite-strings -Wcast-qual -Wstrict-prototypes
-Wmissing-prototypes -Wold-style-definition  -isystem ./include  -fPIC
-mno-minimal-toc -g -DHAVE_GTHR_DEFAULT -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED 
-mlong-double-128 -I. -I. -I../.././gcc -I/home/meissner/fsf-src/trunk/libgcc
-I/home/meissner/fsf-src/trunk/libgcc/.
-I/home/meissner/fsf-src/trunk/libgcc/../gcc
-I/home/meissner/fsf-src/trunk/libgcc/../include
-I/home/meissner/fsf-src/trunk/libgcc/../libdecnumber/dpd
-I/home/meissner/fsf-src/trunk/libgcc/../libdecnumber -DHAVE_CC_TLS  -o
unwind-dw2.o -MT unwind-dw2.o -MD -MP -MF unwind-dw2.dep -fexceptions -c
/home/meissner/fsf-src/trunk/libgcc/../gcc/unwind-dw2.c -fvisibility=hidden
-DHIDE_EXPORTS
In file included from
/home/meissner/fsf-src/trunk/libgcc/../gcc/unwind-dw2.c:1582:0:
/home/meissner/fsf-src/trunk/libgcc/../gcc/unwind.inc: In function
‘_Unwind_RaiseException’:
/home/meissner/fsf-src/trunk/libgcc/../gcc/unwind.inc:136:1: error: insn does
not satisfy its constraints:
(insn 217 216 218 2 (set (reg:SI 11 11)
        (xor:SI (reg:SI 11 11)
            (const_int 3896573952 [0xe8410000])))
/home/meissner/fsf-src/trunk/libgcc/../gcc/unwind.inc:83 158
{*boolsi3_internal1}
     (nil))
/home/meissner/fsf-src/trunk/libgcc/../gcc/unwind.inc:136:1: internal compiler
error: in copyprop_hardreg_forward_1, at regcprop.c:768
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
make[2]: *** [unwind-dw2.o] Error 1
make[2]: Leaving directory
`/home/meissner/fsf-build-ppc64/trunk/powerpc64-linux/libgcc'
make[1]: *** [all-target-libgcc] Error 2
make[1]: Leaving directory `/home/meissner/fsf-build-ppc64/trunk'
make: *** [all] Error 2

[Bug target/45807] Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[meissner at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

--- Comment #5 from Michael Meissner <meissner at gcc dot gnu.org> 2010-10-01 01:51:43 UTC ---
This patch fixes the problem with linux ppc64 builds when the compiler is
defaulting to 64-bit cpus.

This patch is ok to check in.

[Bug target/45807] Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[amodra at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

--- Comment #6 from Alan Modra <amodra at gcc dot gnu.org> 2010-10-01 03:23:50 UTC ---
Author: amodra
Date: Fri Oct  1 03:23:46 2010
New Revision: 164825

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=164825
Log:
    PR target/45807
    * config/rs6000/rs6000.c (rs6000_emit_prologue): Properly sign
    extend toc_restore_insn.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/rs6000/rs6000.c

[Bug target/45807] Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[amodra at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

--- Comment #7 from Alan Modra <amodra at gcc dot gnu.org> 2010-11-17 06:09:58 UTC ---
Author: amodra
Date: Wed Nov 17 06:09:53 2010
New Revision: 166857

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166857
Log:
    PR target/45807
    Backport r164685,r164825,r165101
    * config/rs6000/aix.h (SETUP_FRAME_ADDRESSES): Delete.
    * config/rs6000/linux64.h (SETUP_FRAME_ADDRESSES): Delete.
    * config/rs6000/rs6000-protos.h (rs6000_aix_emit_builtin_unwind_init):
    Delete.
    * config/rs6000/rs6000.c (rs6000_aix_emit_builtin_unwind_init): Delete.
    (rs6000_emit_prologue): Don't just create frame save info for r2,
    actually save r2.


Modified:
    branches/gcc-4_5-branch/gcc/ChangeLog
    branches/gcc-4_5-branch/gcc/config/rs6000/aix.h
    branches/gcc-4_5-branch/gcc/config/rs6000/linux64.h
    branches/gcc-4_5-branch/gcc/config/rs6000/rs6000-protos.h
    branches/gcc-4_5-branch/gcc/config/rs6000/rs6000.c

[Bug target/45807] Lying eh_frame r2 save info causes crashes with static libgcc_eh and libstdc++

[amodra at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45807

Alan Modra <amodra at gmail dot com> changed:

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

--- Comment #8 from Alan Modra <amodra at gmail dot com> 2011-11-07 08:07:10 UTC ---
fixed