[Bug objc/36610] New: message forwarding broken on x86_64: self is not receiver

[Bug objc/36610] New: message forwarding broken on x86_64: self is not receiver

[hydrologiccycle at gmail dot com]

Note that this is a 64-bit problem.  i386 works fine.  Code for reproducing is
included, as well as all system information.


correct output of program on 32-bit:

forwarding for selector display from 134536384 to 134536136
worked, self == receiver
foo is 1234567890, and should be 1234567890


incorrect output of program on 64-bit:

forwarding for selector display from 6330720 to 6330224
broken, self != receiver
foo is 6330224, and should be 1234567890


test program:

// Objective-C x86_64 bug:  self is wrong on forward;
// broken on gcc-4.1.2, 4.3.0, and 4.3.1

#include <stdio.h>
#include <stdlib.h>
#include <objc/Object.h>
#include <objc/objc-api.h>

id forwarder, receiver;

@interface Forwarder:Object
{
    id receiver;
}

-initWithReceiver:theReceiver;

@end

@interface Receiver:Object
{
    int foo;
}

-display;
-initWithFoo:(int)theFoo;

@end

@implementation Receiver

-initWithFoo:(int)theFoo
{
    foo = theFoo;
    return self;
}

-display
{
    if (self == receiver) {
        printf("worked, self == receiver\n");
    } else {
        printf("broken, self != receiver\n");
    }
    printf("foo is %d, and should be %d\n", foo, 1234567890);
    return self;
}

@end

@implementation Forwarder

-initWithReceiver:theReceiver
{
    [super init];
    receiver = theReceiver;
    return self;
}

-(retval_t) forward:(SEL)theSel:(arglist_t)theArgFrame
{
    if (receiver) {

        printf("forwarding for selector %s from %ld to %ld\n",
            sel_get_name(theSel), self, receiver);

        //return objc_msg_sendv(itsNextLink, theSel, theArgFrame);
        return [receiver performv:theSel:theArgFrame];

    } else {
        return [self doesNotRecognize:theSel];
    }
}

@end

int main()
{
    receiver = [[Receiver alloc] initWithFoo:1234567890];
    forwarder = [[Forwarder alloc] initWithReceiver:receiver];

    [forwarder display];

    exit(EXIT_SUCCESS);
    return 0;
}


I compiled gcc from the 4.3.1 tarball on ftp.gnu.org in order to avoid any
distro patches.  Here's the output of gcc -v:

Using built-in specs.
Target: x86_64-pc-linux-gnu
Configured with: ../gcc-4.3.1/configure --prefix=/usr/local
--host=x86_64-pc-linux-gnu --build=x86_64-pc-linux-gnu --disable-altivec
--enable-nls -
-without-included-gettext --with-system-zlib --disable-checking
--disable-werror --enable-secureplt --disable-multilib --enable-libmudflap
--disabl
e-libssp --disable-libgcj --enable-languages=c,objc --enable-shared
--enable-threads=posix --enable-__cxa_atexit --enable-clocale=gnu             
Thread model: posix
gcc version 4.3.1 (GCC)


-- 
           Summary: message forwarding broken on x86_64: self is not
                    receiver
           Product: gcc
           Version: 4.3.1
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: objc
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: hydrologiccycle at gmail dot com
 GCC build triplet: x86_64-pc-linux-gnu
  GCC host triplet: x86_64-pc-linux-gnu
GCC target triplet: x86_64-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] message forwarding broken on x86_64: self is not receiver

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2008-06-24 01:52 -------
This is most likely a problem with __builtin_apply :).


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|major                       |normal
          Component|objc                        |libobjc


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] message forwarding broken on x86_64: self is not receiver

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2008-12-29 02:18 -------
Subject: Bug 36610

Author: pinskia
Date: Mon Dec 29 02:16:45 2008
New Revision: 142945

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142945
Log:
2008-12-28  Andrew Pinski  <pinskia@gmail.com>

        PR libobjc/36610
        * objc/execute/forward-1.m: New test.


Added:
    trunk/gcc/testsuite/objc/execute/forward-1.m
Modified:
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] message forwarding broken on x86_64: self is not receiver

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2009-01-01 16:21 -------
*** Bug 38689 has been marked as a duplicate of this bug. ***


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl dot tools at gmail dot
                   |                            |com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug middle-end/36610] __builtin_apply is broken for targets which pass argument via registers

[pinskia at gcc dot gnu dot org]

------- Comment #4 from pinskia at gcc dot gnu dot org  2009-01-01 16:23 -------
Confirmed, this is really a __builtin_apply bug.  Libobjc either needs to move
to libffi or __builtin_apply needs to be fixed up.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
          Component|libobjc                     |middle-end
     Ever Confirmed|0                           |1
  GCC build triplet|x86_64-pc-linux-gnu         |
   GCC host triplet|x86_64-pc-linux-gnu         |
 GCC target triplet|x86_64-pc-linux-gnu         |arguments in registers
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2009-01-01 16:23:10
               date|                            |
            Summary|message forwarding broken on|__builtin_apply is broken
                   |x86_64: self is not receiver|for targets which pass
                   |                            |argument via registers


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug middle-end/36610] __builtin_apply is broken for targets which pass argument via registers

[hjl dot tools at gmail dot com]

------- Comment #5 from hjl dot tools at gmail dot com  2009-01-01 16:26 -------
(In reply to comment #4)
> Confirmed, this is really a __builtin_apply bug.  Libobjc either needs to move
> to libffi or __builtin_apply needs to be fixed up.
> 

Is there a __builtin_apply testcase in C?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug objc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[matz at gcc dot gnu dot org]

------- Comment #6 from matz at gcc dot gnu dot org  2009-02-12 11:32 -------
I don't think it's __builtin_apply which is wrong.  It's rather how it is
used in libobjc.  In particular register parameters are handled wrongly.
libobjc objc_msg_sendv() simply tries to overwrite the first two argument
slots returned by __builtin_apply_args (called in a different routine) with
those it really wants in there.  It uses method_get_{first,next}_argument
for that which tries to use the argument pointer in that memory block.

But memory pointed to by that arg pointer only contains the args passed on 
stack.  Those passed in registers lie somewhere else (in the block returned
by builtin_apply_args, but behind the arg pointer), in register order (not
in argument order!).

So, overwriting the argument slots doesn't actually overwrite the data which
later is used in __builtin_apply --> boom.

Accordingly changing the summary.


-- 

matz at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |objc
            Summary|__builtin_apply is broken   |objc_msg_sendv is broken for
                   |for targets which pass      |targets which pass argument
                   |argument via registers      |via registers


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug objc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[ghazi at gcc dot gnu dot org]

------- Comment #7 from ghazi at gcc dot gnu dot org  2009-03-27 16:51 -------
Andrew - You added the testcase for this PR back in December to see which lp64
targets failed:
http://gcc.gnu.org/ml/gcc-patches/2008-12/msg01199.html

It's still failing several months later.
x86_64: http://gcc.gnu.org/ml/gcc-patches/2008-12/msg01199.html
ppc64: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02732.html
ia64: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02716.html
s390: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02491.html
sh4: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02437.html

etc.

I think there's plenty of data now to see where it fails.  You should XFAIL it
until a fix is installed to avoid noise in the testsuite results.

Thanks.


-- 

ghazi at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ghazi at gcc dot gnu dot
                   |                            |org, pinskia at gcc dot gnu
                   |                            |dot org
   Last reconfirmed|2009-01-01 16:23:10         |2009-03-27 16:51:48
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug objc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[ghazi at gcc dot gnu dot org]

------- Comment #8 from ghazi at gcc dot gnu dot org  2009-03-27 16:55 -------
(In reply to comment #7)
> It's still failing several months later.
> x86_64: http://gcc.gnu.org/ml/gcc-patches/2008-12/msg01199.html

The x86_64 link should be:
http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02744.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug objc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[dave at hiauly1 dot hia dot nrc dot ca]

------- Comment #9 from dave at hiauly1 dot hia dot nrc dot ca  2009-03-27 17:38 -------
Subject: Re:  objc_msg_sendv is broken for targets which pass argument via
registers

> It's still failing several months later.
> x86_64: http://gcc.gnu.org/ml/gcc-patches/2008-12/msg01199.html
> ppc64: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02732.html
> ia64: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02716.html
> s390: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02491.html
> sh4: http://gcc.gnu.org/ml/gcc-testresults/2009-03/msg02437.html

Also, hppa.

> I think there's plenty of data now to see where it fails.  You should XFAIL it
> until a fix is installed to avoid noise in the testsuite results.

I tried but .x files don't seem to work in this directory.

Dave


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug objc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[ghazi at gcc dot gnu dot org]

------- Comment #10 from ghazi at gcc dot gnu dot org  2009-03-31 19:12 -------
(In reply to comment #9)
> > I think there's plenty of data now to see where it fails.  You should XFAIL it
> > until a fix is installed to avoid noise in the testsuite results.
> I tried but .x files don't seem to work in this directory.
> Dave

I can't get dg- style to work either.  What I'd like to do is something like
this:

/* { dg-xfail-run-if "PR36610" { lp64 } "*" "" } */

Janis, can you please help?  Thanks.


-- 

ghazi at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |janis at gcc dot gnu dot org
   Last reconfirmed|2009-03-27 16:51:48         |2009-03-31 19:12:03
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[pinskia at gcc dot gnu dot org]

------- Comment #11 from pinskia at gcc dot gnu dot org  2009-03-31 19:34 -------
(In reply to comment #10)
> I can't get dg- style to work either.  What I'd like to do is something like
> this:
> 
> /* { dg-xfail-run-if "PR36610" { lp64 } "*" "" } */
> 
> Janis, can you please help?  Thanks.

Except that is not a fully correct xfail as it is more than just lp64 targets,
it is about targets which pass via registers.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|objc                        |libobjc


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[janis at gcc dot gnu dot org]

------- Comment #12 from janis at gcc dot gnu dot org  2009-03-31 20:28 -------
The objc/execute directory can process .x files like the one added here, which
I've tested on powerpc64-linux with -m32/-m64:

Index: gcc/testsuite/objc/execute/forward-1.x
===================================================================
--- gcc/testsuite/objc/execute/forward-1.x      (revision 0)
+++ gcc/testsuite/objc/execute/forward-1.x      (revision 0)
@@ -0,0 +1,8 @@
+load_lib target-supports.exp
+
+if { ([istarget x86_64-*-linux*] && [is_effective_target_lp64])
+     || [istarget powerpc*-*-linux*] } {
+    set torture_execute_xfail "*-*-*"
+}
+
+return 0

As Andrew said, the list of affected targets is longer than this.  I don't
think it's worth adding an effective-target keyword, people can just add new
targets to this list.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[janis at gcc dot gnu dot org]

------- Comment #13 from janis at gcc dot gnu dot org  2009-04-09 16:58 -------
Subject: Bug 36610

Author: janis
Date: Thu Apr  9 16:58:34 2009
New Revision: 145849

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=145849
Log:
        PR libobjc/36610
        * objc/execute/forward-1.x: New.

Added:
    trunk/gcc/testsuite/objc/execute/forward-1.x
Modified:
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[janis at gcc dot gnu dot org]

------- Comment #14 from janis at gcc dot gnu dot org  2009-04-09 17:01 -------
Subject: Bug 36610

Author: janis
Date: Thu Apr  9 17:00:57 2009
New Revision: 145850

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=145850
Log:
        PR libobjc/36610
        * objc/execute/forward-1.x: New.

Added:
    branches/gcc-4_4-branch/gcc/testsuite/objc/execute/forward-1.x
Modified:
    branches/gcc-4_4-branch/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[pinskia at gcc dot gnu dot org]

------- Comment #15 from pinskia at gcc dot gnu dot org  2009-04-19 15:52 -------
*** Bug 39817 has been marked as a duplicate of this bug. ***


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |js-gcc at webkeks dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[ro at gcc dot gnu dot org]

------- Comment #16 from ro at gcc dot gnu dot org  2010-02-24 11:56 -------
Subject: Bug 36610

Author: ro
Date: Wed Feb 24 11:56:10 2010
New Revision: 157035

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=157035
Log:
        PR libobjc/36610
        * objc/execute/forward-1.x: XFAIL on alpha*-dec-osf*, 64-bit
        i?86-*-solaris2*, mips-sgi-irix*, sparc*-sun-solaris2* with
        -fgnu-runtime.
        Sort entries.

Modified:
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/objc/execute/forward-1.x


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610

[Bug libobjc/36610] objc_msg_sendv is broken for targets which pass argument via registers

[ro at gcc dot gnu dot org]

------- Comment #17 from ro at gcc dot gnu dot org  2010-06-02 17:17 -------
Subject: Bug 36610

Author: ro
Date: Wed Jun  2 17:16:55 2010
New Revision: 160172

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=160172
Log:
        Backport from mainline:
        2010-02-24  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

        PR libobjc/36610
        * objc/execute/forward-1.x: XFAIL on alpha*-dec-osf*, 64-bit
        i?86-*-solaris2*, mips-sgi-irix*, sparc*-sun-solaris2* with
        -fgnu-runtime.
        Sort entries.

Modified:
    branches/gcc-4_4-branch/gcc/testsuite/ChangeLog
    branches/gcc-4_4-branch/gcc/testsuite/objc/execute/forward-1.x


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36610