Re: Memory Violation With Windres and Shared Data Segments

Re: Memory Violation With Windres and Shared Data Segments

[Dirk Fieldhouse]

On Sat, 07 Aug 1999 22:55:16 -0500, Mumit Khan <khan@xraylith.wisc.EDU> wrote:
>...>[binutils &c enhancements for shareable data segments]

>Just to let you folks know, I just submitted patches to both binutils >and gcc to make this happen.
>
>In the future, you'll be able to do this:
> 
>  int __attribute__((section ("shared"), shared)) foo = 0;
>  int
>  main ()
>  {
>    /* do stuff to foo. */
>    return 0;
>  }
>
>and all running copies of the executable will share ``foo''.
>...

That certainly explains why my hook DLL with 
 
static HHOOK __attribute__ ((section("SHARED"))) g_hHook = NULL;

(hoping for magic in the section name) didn't fly!

I'll look forward to a better result with the enhanced tools.

Thanks
/df

-- 
Dirk Fieldhouse                 Logica UK Limited
fieldhouse@logica.com           75 Hampstead Road
c=gb;a=attmail;p=logica;        London NW1 2PL
o=LOGICA;ou1=UK;s=fieldhouse    UK
+44 (171) 637 9111

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Dirk Fieldhouse]

On Sat, 07 Aug 1999 22:55:16 -0500, Mumit Khan <khan@xraylith.wisc.EDU> wrote:
>...>[binutils &c enhancements for shareable data segments]

>Just to let you folks know, I just submitted patches to both binutils >and gcc to make this happen.
>
>In the future, you'll be able to do this:
> 
>  int __attribute__((section ("shared"), shared)) foo = 0;
>  int
>  main ()
>  {
>    /* do stuff to foo. */
>    return 0;
>  }
>
>and all running copies of the executable will share ``foo''.
>...

That certainly explains why my hook DLL with 
 
static HHOOK __attribute__ ((section("SHARED"))) g_hHook = NULL;

(hoping for magic in the section name) didn't fly!

I'll look forward to a better result with the enhanced tools.

Thanks
/df

-- 
Dirk Fieldhouse                 Logica UK Limited
fieldhouse@logica.com           75 Hampstead Road
c=gb;a=attmail;p=logica;        London NW1 2PL
o=LOGICA;ou1=UK;s=fieldhouse    UK
+44 (171) 637 9111

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Harold Weissfield]

>I believe the only difference is that it has the "SHARED" bit set in
>the section attributes, in addition to the usual initialized data,
>read-write, etc flags.
>
>Can someone send me a testcase so I can see if my local changes fix this
>or not? If it's obvious already, I know nothing about these shared
>segments.
>
>Regards,
>Mumit

Thank you for your help, Mumit, Chris, and Anders.  I will try to experiment 
around with the information that you guys have provided, and I shall reply 
back if I find any favorable results.

In case you are wondering why I'm doing this, I am writing a shell 
replacement for Explorer for our lab at school, and one of the 
implementation details involves creating a global hook using 
SetWindowsHookEx().  Unfortunately, the requirements for having such a hook 
involve having a shared data segment in a DLL, so that is the dilema that I 
am presently trying to solve.

Mumit, I am not aware of the proper procedures for submitting code, so 
forgive me if my approach is mistaken.  During the process of developing 
this shell, I noticed that the following notification messages for the 
appbar were not in shellapi.h

#define ABN_STATECHANGE    0x0000000
#define ABN_POSCHANGED     0x0000001
#define ABN_FULLSCREENAPP  0x0000002
#define ABN_WINDOWARRANGE  0x0000003

Please feel free to add them to the headers at your convenience.  Meanwhile, 
I thank you once again, and I shall get busy on my coding.

Cheers,
Harold Weissfield


_______________________________________________________________
Get Free Email and Do More On The Web. Visit http://www.msn.com

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

Anders Norlander <anorland@hem2.passagen.se> writes:
> Just naming a section shared doesn't actually make it shared. There is a
> bit that
> has to be set in order to make a section shared. Unfortunately I don't
> remember
> more right now.

I believe the only difference is that it has the "SHARED" bit set in 
the section attributes, in addition to the usual initialized data,
read-write, etc flags.

Can someone send me a testcase so I can see if my local changes fix this
or not? If it's obvious already, I know nothing about these shared 
segments.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

"Harold Weissfield" <h_weissfield@hotmail.com> writes:
> 
> Thank you for your help, Mumit, Chris, and Anders.  I will try to experiment 
> around with the information that you guys have provided, and I shall reply 
> back if I find any favorable results.

I did look at it a bit, and here's what's needs to be done:
  
  - bfd (binutils): Need to be able to specify a shared flag for sections.
    That part is somewhat easy.
  - gas (binutils): Need to be able parse section flags containing shared
    flag. Also easy.
  - gcc: Somehow we need to be able to tell the compiler about this "shared"
    link. MSVC does it with a linker comment, but GCC doesn't have such
    a facility directly. It's going to take a bit of work. I might add
    an __attribute__((shared)), but haven't thought of the other
    possibilities yet.

The current binutils release simply won't work -- the CVS/development tree
has some fixes that are absolutely needed for this to even start working.
However, my quick test is somehow creating bad executable, so more work
needs to be done.

> Mumit, I am not aware of the proper procedures for submitting code, so 
> forgive me if my approach is mistaken.  During the process of developing 
> this shell, I noticed that the following notification messages for the 
> appbar were not in shellapi.h
> 
> #define ABN_STATECHANGE    0x0000000
> #define ABN_POSCHANGED     0x0000001
> #define ABN_FULLSCREENAPP  0x0000002
> #define ABN_WINDOWARRANGE  0x0000003
> 
> Please feel free to add them to the headers at your convenience.  Meanwhile, 
> I thank you once again, and I shall get busy on my coding.

This is Anders' baby. Since he's on the list, we're safe.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

Chris Faylor <cgf@cygnus.com> writes:
> On Fri, Aug 06, 1999 at 02:57:23PM +0000, Harold Weissfield wrote:
> >On a different note, I have been searching through the documentation for GCC
>  
> >for information on how to created a shared data segment in a DLL, but so far
>  
> >have been unable to come up with anything.  Under Visual C++, this was done 
> >with a #pragma comment(linker, "-section:Shared,rws") and then a #pragma 
> >data_seg("Shared").  Is there any equivalent construct in GCC?
> I don't think that there is a way to do this currently with gcc/binutils tool
> s,
> unfortunately.  I used to use Visual C tools to make a section sharable when
> I needed to do this.

You can create these named sections, but there is a bug the linker that
causes these sections to be readonly. I fixed this in the development
sources a while back, so the next release will work as expected. However, 
I don't what the exact semantics of this "shared" segment is, but if it's 
just a named section like .tls (thread local storage), then it should work
fine.

The construct is like this:
  
  int foo __attribute__((section("shared"))) = 0;

now the *initialized* variable `foo' goes into a special section `shared'; 
the `shared' section should be marked r/w, but currently GNU binutils will 
mark it readonly due to a tiny bug.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Memory Violation With Windres and Shared Data Segments

[Harold Weissfield]

Gentlemen, I have recently downloaded Mumit's Mingw32 distribution with GCC 
2.95 as the compiler, and everything is working well.  However, I have had 
the consistent problem of when I tried to use windres, Windows pops up a 
messagebox that says something similar to the following:

"Your program accessed memory currently in use at 00003FBC from 1654:02FE.

You should save all your work, and then quit your programs.  If you continue 
running this program, your system may become unstable.  Do you want to 
continue?"

I was wondering if anyone else had encountered this problem before or if it 
was just my computer at fault.

On a different note, I have been searching through the documentation for GCC 
for information on how to created a shared data segment in a DLL, but so far 
have been unable to come up with anything.  Under Visual C++, this was done 
with a #pragma comment(linker, "-section:Shared,rws") and then a #pragma 
data_seg("Shared").  Is there any equivalent construct in GCC?

I appreciate any comments or suggestions, and thank you for your time.

Cheers,
Harold Weissfield


_______________________________________________________________
Get Free Email and Do More On The Web. Visit http://www.msn.com

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

Mumit Khan <khan@xraylith.wisc.EDU> writes:
> I did look at it a bit, and here's what's needs to be done:
>   
>   - bfd (binutils): Need to be able to specify a shared flag for sections.
>     That part is somewhat easy.
>   - gas (binutils): Need to be able parse section flags containing shared
>     flag. Also easy.
>   - gcc: Somehow we need to be able to tell the compiler about this "shared"
>     link. MSVC does it with a linker comment, but GCC doesn't have such
>     a facility directly. It's going to take a bit of work. I might add
>     an __attribute__((shared)), but haven't thought of the other
>     possibilities yet.

Just to let you folks know, I just submitted patches to both binutils and
gcc to make this happen.

In the future, you'll be able to do this:
  
  int __attribute__((section ("shared"), shared)) foo = 0;
  int 
  main ()
  {
    /* do stuff to foo. */
    return 0;
  }

and all running copies of the executable will share ``foo''.

Anyone who wants to test it, let me know. I have updated copies of 
binutils-2.9.4 and gcc-2.95 binaries that you can put somewhere,
and use the -B<prefix>/ argument to test it.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Chris Faylor]

On Fri, Aug 06, 1999 at 02:57:23PM +0000, Harold Weissfield wrote:
>On a different note, I have been searching through the documentation for GCC 
>for information on how to created a shared data segment in a DLL, but so far 
>have been unable to come up with anything.  Under Visual C++, this was done 
>with a #pragma comment(linker, "-section:Shared,rws") and then a #pragma 
>data_seg("Shared").  Is there any equivalent construct in GCC?

I don't think that there is a way to do this currently with gcc/binutils tools,
unfortunately.  I used to use Visual C tools to make a section sharable when
I needed to do this.

-chris

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

Mumit Khan <khan@xraylith.wisc.EDU> writes:
> I did look at it a bit, and here's what's needs to be done:
>   
>   - bfd (binutils): Need to be able to specify a shared flag for sections.
>     That part is somewhat easy.
>   - gas (binutils): Need to be able parse section flags containing shared
>     flag. Also easy.
>   - gcc: Somehow we need to be able to tell the compiler about this "shared"
>     link. MSVC does it with a linker comment, but GCC doesn't have such
>     a facility directly. It's going to take a bit of work. I might add
>     an __attribute__((shared)), but haven't thought of the other
>     possibilities yet.

Just to let you folks know, I just submitted patches to both binutils and
gcc to make this happen.

In the future, you'll be able to do this:
  
  int __attribute__((section ("shared"), shared)) foo = 0;
  int 
  main ()
  {
    /* do stuff to foo. */
    return 0;
  }

and all running copies of the executable will share ``foo''.

Anyone who wants to test it, let me know. I have updated copies of 
binutils-2.9.4 and gcc-2.95 binaries that you can put somewhere,
and use the -B<prefix>/ argument to test it.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

"Harold Weissfield" <h_weissfield@hotmail.com> writes:
> 
> Thank you for your help, Mumit, Chris, and Anders.  I will try to experiment 
> around with the information that you guys have provided, and I shall reply 
> back if I find any favorable results.

I did look at it a bit, and here's what's needs to be done:
  
  - bfd (binutils): Need to be able to specify a shared flag for sections.
    That part is somewhat easy.
  - gas (binutils): Need to be able parse section flags containing shared
    flag. Also easy.
  - gcc: Somehow we need to be able to tell the compiler about this "shared"
    link. MSVC does it with a linker comment, but GCC doesn't have such
    a facility directly. It's going to take a bit of work. I might add
    an __attribute__((shared)), but haven't thought of the other
    possibilities yet.

The current binutils release simply won't work -- the CVS/development tree
has some fixes that are absolutely needed for this to even start working.
However, my quick test is somehow creating bad executable, so more work
needs to be done.

> Mumit, I am not aware of the proper procedures for submitting code, so 
> forgive me if my approach is mistaken.  During the process of developing 
> this shell, I noticed that the following notification messages for the 
> appbar were not in shellapi.h
> 
> #define ABN_STATECHANGE    0x0000000
> #define ABN_POSCHANGED     0x0000001
> #define ABN_FULLSCREENAPP  0x0000002
> #define ABN_WINDOWARRANGE  0x0000003
> 
> Please feel free to add them to the headers at your convenience.  Meanwhile, 
> I thank you once again, and I shall get busy on my coding.

This is Anders' baby. Since he's on the list, we're safe.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Harold Weissfield]

>I believe the only difference is that it has the "SHARED" bit set in
>the section attributes, in addition to the usual initialized data,
>read-write, etc flags.
>
>Can someone send me a testcase so I can see if my local changes fix this
>or not? If it's obvious already, I know nothing about these shared
>segments.
>
>Regards,
>Mumit

Thank you for your help, Mumit, Chris, and Anders.  I will try to experiment 
around with the information that you guys have provided, and I shall reply 
back if I find any favorable results.

In case you are wondering why I'm doing this, I am writing a shell 
replacement for Explorer for our lab at school, and one of the 
implementation details involves creating a global hook using 
SetWindowsHookEx().  Unfortunately, the requirements for having such a hook 
involve having a shared data segment in a DLL, so that is the dilema that I 
am presently trying to solve.

Mumit, I am not aware of the proper procedures for submitting code, so 
forgive me if my approach is mistaken.  During the process of developing 
this shell, I noticed that the following notification messages for the 
appbar were not in shellapi.h

#define ABN_STATECHANGE    0x0000000
#define ABN_POSCHANGED     0x0000001
#define ABN_FULLSCREENAPP  0x0000002
#define ABN_WINDOWARRANGE  0x0000003

Please feel free to add them to the headers at your convenience.  Meanwhile, 
I thank you once again, and I shall get busy on my coding.

Cheers,
Harold Weissfield


_______________________________________________________________
Get Free Email and Do More On The Web. Visit http://www.msn.com

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

Anders Norlander <anorland@hem2.passagen.se> writes:
> Just naming a section shared doesn't actually make it shared. There is a
> bit that
> has to be set in order to make a section shared. Unfortunately I don't
> remember
> more right now.

I believe the only difference is that it has the "SHARED" bit set in 
the section attributes, in addition to the usual initialized data,
read-write, etc flags.

Can someone send me a testcase so I can see if my local changes fix this
or not? If it's obvious already, I know nothing about these shared 
segments.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Mumit Khan]

Chris Faylor <cgf@cygnus.com> writes:
> On Fri, Aug 06, 1999 at 02:57:23PM +0000, Harold Weissfield wrote:
> >On a different note, I have been searching through the documentation for GCC
>  
> >for information on how to created a shared data segment in a DLL, but so far
>  
> >have been unable to come up with anything.  Under Visual C++, this was done 
> >with a #pragma comment(linker, "-section:Shared,rws") and then a #pragma 
> >data_seg("Shared").  Is there any equivalent construct in GCC?
> I don't think that there is a way to do this currently with gcc/binutils tool
> s,
> unfortunately.  I used to use Visual C tools to make a section sharable when
> I needed to do this.

You can create these named sections, but there is a bug the linker that
causes these sections to be readonly. I fixed this in the development
sources a while back, so the next release will work as expected. However, 
I don't what the exact semantics of this "shared" segment is, but if it's 
just a named section like .tls (thread local storage), then it should work
fine.

The construct is like this:
  
  int foo __attribute__((section("shared"))) = 0;

now the *initialized* variable `foo' goes into a special section `shared'; 
the `shared' section should be marked r/w, but currently GNU binutils will 
mark it readonly due to a tiny bug.

Regards,
Mumit


--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Re: Memory Violation With Windres and Shared Data Segments

[Chris Faylor]

On Fri, Aug 06, 1999 at 02:57:23PM +0000, Harold Weissfield wrote:
>On a different note, I have been searching through the documentation for GCC 
>for information on how to created a shared data segment in a DLL, but so far 
>have been unable to come up with anything.  Under Visual C++, this was done 
>with a #pragma comment(linker, "-section:Shared,rws") and then a #pragma 
>data_seg("Shared").  Is there any equivalent construct in GCC?

I don't think that there is a way to do this currently with gcc/binutils tools,
unfortunately.  I used to use Visual C tools to make a section sharable when
I needed to do this.

-chris

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com

Memory Violation With Windres and Shared Data Segments

[Harold Weissfield]

Gentlemen, I have recently downloaded Mumit's Mingw32 distribution with GCC 
2.95 as the compiler, and everything is working well.  However, I have had 
the consistent problem of when I tried to use windres, Windows pops up a 
messagebox that says something similar to the following:

"Your program accessed memory currently in use at 00003FBC from 1654:02FE.

You should save all your work, and then quit your programs.  If you continue 
running this program, your system may become unstable.  Do you want to 
continue?"

I was wondering if anyone else had encountered this problem before or if it 
was just my computer at fault.

On a different note, I have been searching through the documentation for GCC 
for information on how to created a shared data segment in a DLL, but so far 
have been unable to come up with anything.  Under Visual C++, this was done 
with a #pragma comment(linker, "-section:Shared,rws") and then a #pragma 
data_seg("Shared").  Is there any equivalent construct in GCC?

I appreciate any comments or suggestions, and thank you for your time.

Cheers,
Harold Weissfield


_______________________________________________________________
Get Free Email and Do More On The Web. Visit http://www.msn.com

--
Want to unsubscribe from this list?
Send a message to cygwin-unsubscribe@sourceware.cygnus.com