Re: [Qemu-devel] [PATCH v2 05/39] fdsets: use weak aliases instead of qemu-tool.c/qemu-user.c

[Stefan Weil]

Am 15.11.2012 21:52, schrieb Paolo Bonzini:
> Il 15/11/2012 19:01, Stefan Weil ha scritto:
> > Hi Paolo,
> >
> > this patch breaks QEMU on 32 and 64 bit hosts, native and with Wine.
> > It's easy to reproduce the SIGSEGV crash: just add a -snapshot option.
> > Obviously the critical code is executed only when this option was used.
>
> I cannot reproduce this, so it must be an assembler or linker bug.
>
> Can you try the alternative code that is used for Mac OS X?
>
> Paolo

The code which is used for Mac OS X also compiles and
results in the same run-time bug with Wine:

wine: Unhandled page fault on write access to 0x00000004 at address 
0x7b845d6e (thread 001b), starting debugger...

(immediately after BIOS says "Booting from hard disk...")

This was the modification used:

diff --git a/compiler.h b/compiler.h
index 55d7d74..62427e4 100644
--- a/compiler.h
+++ b/compiler.h
@@ -50,11 +50,12 @@
 #   define __printf__ __gnu_printf__
 #  endif
 # endif
-# if defined(__APPLE__)
+# if defined(__APPLE__) || defined(_WIN32)
 #  define QEMU_WEAK_ALIAS(newname, oldname) \
         static typeof(oldname) weak_##newname __attribute__((unused, 
weakref(#oldname)))
 #  define QEMU_WEAK_REF(newname, oldname) (weak_##newname ? 
weak_##newname : oldname)
 # else
+#error
 #  define QEMU_WEAK_ALIAS(newname, oldname) \
         typeof(oldname) newname __attribute__((weak, alias (#oldname)))
 #  define QEMU_WEAK_REF(newname, oldname) newname


These are my Debian packages (only the 32 bit ones are needed for the test):

ii  binutils-mingw-w64-i686                     
2.22-7+2                           Cross-binutils for Win32 (x86) using 
MinGW-w64
ii  binutils-mingw-w64-x86-64                   
2.22-7+2                           Cross-binutils for Win64 (x64) using 
MinGW-w64
ii  gcc-mingw-w64                               
4.6.3-8+7                          GNU C compiler for MinGW-w64
ii  gcc-mingw-w64-base                          
4.6.3-8+7                          GNU Compiler Collection for MinGW-w64 
(base package)
ii  gcc-mingw-w64-i686                          
4.6.3-8+7                          GNU C compiler for MinGW-w64 
targeting Win32
ii  gcc-mingw-w64-x86-64                        
4.6.3-8+7                          GNU C compiler for MinGW-w64 
targeting Win64
ii  mingw-w64                                   
3.0~svn4933-1                      Development environment targetting 
32- and 64-bit Windows
ii  mingw-w64-dev                               
3.0~svn4933-1                      Development files for MinGW-w64
ii  mingw-w64-tools                             
3.0~svn4933-1                      Development tools for 32- and 64-bit 
Windows

On Windows, I used a rather new MinGW standard installation.

I'll run more tests with other Linux distributions tomorrow.

Cheers,

Stefan


> > Here is a simple command line using Wine:
> >
> > wine i386-softmmu/qemu-system-i386 -L pc-bios -snapshot Makefile
> >
> > The disk image does not matter, so I just selected QEMU's Makefile.
> >
> > It looks like weak symbols are not really working with MinGW
> > (Blue Swirl previously pointed out that only ELF and a.out are
> > officially supported).
> >
> > I can see in the debugger that QEMU wants to call monitor_fdset_dup_fd_find
> > from qemu_close.
> >
> > In previous versions, this was just a dummy function returning 0.
> > Now, it is the function in monitor.c, but the address does not match
> > exactly, so the code addresses lines near the beginning of
> > monitor_fdset_dup_fd_find which does not work of course.
> >
> > A trivial workaround is calling default_fdset_dup_fd_find which
> > restores the old behaviour. I expect that all other weak functions
> > would show the same problem if they were used.
> >
> > Regards,
> >
> > Stefan