Memory problem in winelib apps?

[Walt Ogburn]

Hi developers,

I'm trying to understand why the following simple program crashes.  I
think this is the same problem that prevents jack_fst from running with
the current version of wine.  (Jack_fst allows a lot of programs called
VSTs to be run, for audio work).

I compiled this with:
> winemaker .
> make

and ran it with:
> wine membug.exe.so

If BUFSIZE is 1044097, there's an error like this:
err:seh:setup_exception stack overflow 12 bytes in thread 0009 eip 4057426f esp 40580ff4 stack 0x40580000-0x40680000

If BUFSIZE is 1044096 or lower, there's no error.  If BUFSIZE is much
higher (like 2000000), there's a segfault.

All the program does is define a character array of size BUFSIZE and
scribble in it to make sure it really gets allocated.  Maybe this simple
example will show how to make jack_fst work.

Thanks,
Walter

[Mike McCormack]

Walt Ogburn wrote:
> #define BUFSIZE 1044096
> /* #define BUFSIZE 2000000 */
>
> int PASCAL WinMain (HINSTANCE inst, HINSTANCE prev, LPSTR cmdline, int show)
> {
>    char buf[BUFSIZE];
>    int i;

Wine allocates a 1Mb stack by default, and more if a larger stack size 
is specified in a PE executeable's header.  You can specify the size of 
the a .exe.so's stack by making a def file for it, and adding the line

STACKSIZE 3000000

that should fix the problem, however allocating a 2M buffer on the stack 
seems like a waste of stack space, so it would be better to fix the 
program to allocate memory on the heap, or directly using mmap.

Mike

[Walt Ogburn]

Thanks Mike.  I think the fixed problem Chris mentioned was a different
one, so I'll see if I can make jack_fst work properly using your
suggestion.

- Walter


On Wed, 22 Jun 2005, Mike McCormack wrote:
> Walt Ogburn wrote:
> > #define BUFSIZE 1044096
> > /* #define BUFSIZE 2000000 */
> >
> > int PASCAL WinMain (HINSTANCE inst, HINSTANCE prev, LPSTR cmdline, int show)
> > {
> >    char buf[BUFSIZE];
> >    int i;
>
> Wine allocates a 1Mb stack by default, and more if a larger stack size
> is specified in a PE executeable's header.  You can specify the size of
> the a .exe.so's stack by making a def file for it, and adding the line
>
> STACKSIZE 3000000
>
> that should fix the problem, however allocating a 2M buffer on the stack
> seems like a waste of stack space, so it would be better to fix the
> program to allocate memory on the heap, or directly using mmap.
>
> Mike

[Alex Villací­s Lasso]

Walt Ogburn wrote:
> Hi developers,
>
> I'm trying to understand why the following simple program crashes.  I
> think this is the same problem that prevents jack_fst from running with
> the current version of wine.  (Jack_fst allows a lot of programs called
> VSTs to be run, for audio work).
>
> I compiled this with:
> > winemaker .
> > make
>
> and ran it with:
> > wine membug.exe.so
>
> If BUFSIZE is 1044097, there's an error like this:
> err:seh:setup_exception stack overflow 12 bytes in thread 0009 eip 4057426f esp 40580ff4 stack 0x40580000-0x40680000
>
> If BUFSIZE is 1044096 or lower, there's no error.  If BUFSIZE is much
> higher (like 2000000), there's a segfault.
>
> All the program does is define a character array of size BUFSIZE and
> scribble in it to make sure it really gets allocated.  Maybe this simple
> example will show how to make jack_fst work.
>
> Thanks,
> Walter
>
> ------------------------------------------------------------------------
>
> #include "windows.h"
>
> #define BUFSIZE 1044096
> /* #define BUFSIZE 2000000 */
>
> int PASCAL WinMain (HINSTANCE inst, HINSTANCE prev, LPSTR cmdline, int show)
> {
>   char buf[BUFSIZE];
>   int i;
>
> for (i=0; i<BUFSIZE; i++)
>   {
>      buf[i] = (char) (i % 256);
>   }
>
> return 0;
> }

Your sample program is trying to use almost 1 megabyte (!) of stack 
space. No wonder it crashes. Are you sure that your program jack_fst is 
trying to allocate such an enormous amount of stack space? Most (sane) 
programs would just declare a pointer and allocate the required space 
via malloc() or HeapAlloc().

Where can somebody else get a copy of jack_fst? Is there a public 
download somewhere?
Maybe jack_fst corrupts the stack in some strange way that was 
misdiagnosed as as request for a 1Mb stack?

[Chris Morgan]

This should have been fixed in jack around 20040325 with a jack to the jack 
variable, JACK_THREAD_STACK_TOUCH.  The value was changed from 1048576
(#define BIG_ENOUGH_STACK in client.c:jack_activate():1033 or so) and moved 
into configure.in.

The change was made in this rev of configure.in:
----------------------------
revision 1.253
date: 2004/03/26 04:41:57;  author: joq;  state: Exp;  lines: +17 -4
[0.96.1] config fixes for shm; stack touch
----------------------------

The reason jack does this is to ensure that all pages of the shared memory for 
all jack clients is mapped so there are no delays during excution.  The value 
was reduced because wine was taking up a bunch of the stack and jack/wine 
were crashing.

Which version of jack are you using?  Which version of wine?

The last time I tried jack under wine it didn't work but I wasn't able to 
figure out why this time and haven't gone back to it in a few months.

Chris




On Tuesday 21 June 2005 10:02 pm, Alex Villací­s Lasso wrote:
> Walt Ogburn wrote:
> > Hi developers,
> >
> > I'm trying to understand why the following simple program crashes.  I
> > think this is the same problem that prevents jack_fst from running with
> > the current version of wine.  (Jack_fst allows a lot of programs called
> > VSTs to be run, for audio work).
> >
> > I compiled this with:
> > > winemaker .
> > > make
> >
> > and ran it with:
> > > wine membug.exe.so
> >
> > If BUFSIZE is 1044097, there's an error like this:
> > err:seh:setup_exception stack overflow 12 bytes in thread 0009 eip
> > 4057426f esp 40580ff4 stack 0x40580000-0x40680000
> >
> > If BUFSIZE is 1044096 or lower, there's no error.  If BUFSIZE is much
> > higher (like 2000000), there's a segfault.
> >
> > All the program does is define a character array of size BUFSIZE and
> > scribble in it to make sure it really gets allocated.  Maybe this simple
> > example will show how to make jack_fst work.
> >
> > Thanks,
> > Walter
> >
> > ------------------------------------------------------------------------
> >
> > #include "windows.h"
> >
> > #define BUFSIZE 1044096
> > /* #define BUFSIZE 2000000 */
> >
> > int PASCAL WinMain (HINSTANCE inst, HINSTANCE prev, LPSTR cmdline, int
> > show) {
> >   char buf[BUFSIZE];
> >   int i;
> >
> > for (i=0; i<BUFSIZE; i++)
> >   {
> >      buf[i] = (char) (i % 256);
> >   }
> >
> > return 0;
> > }
>
> Your sample program is trying to use almost 1 megabyte (!) of stack
> space. No wonder it crashes. Are you sure that your program jack_fst is
> trying to allocate such an enormous amount of stack space? Most (sane)
> programs would just declare a pointer and allocate the required space
> via malloc() or HeapAlloc().
>
> Where can somebody else get a copy of jack_fst? Is there a public
> download somewhere?
> Maybe jack_fst corrupts the stack in some strange way that was
> misdiagnosed as as request for a 1Mb stack?

[Walt Ogburn]

Thanks Chris.  Mark, do you see the problem go away with a later Jack
version?  I was just trying to see if I could understand better what was
going on when we were trying to get this working last year.

- Walter



On Tue, 21 Jun 2005, Chris Morgan wrote:
> This should have been fixed in jack around 20040325 with a jack to the jack
> variable, JACK_THREAD_STACK_TOUCH.  The value was changed from 1048576
> (#define BIG_ENOUGH_STACK in client.c:jack_activate():1033 or so) and moved
> into configure.in.
>
> The change was made in this rev of configure.in:
> ----------------------------
> revision 1.253
> date: 2004/03/26 04:41:57;  author: joq;  state: Exp;  lines: +17 -4
> [0.96.1] config fixes for shm; stack touch
> ----------------------------
>
> The reason jack does this is to ensure that all pages of the shared memory for
> all jack clients is mapped so there are no delays during excution.  The value
> was reduced because wine was taking up a bunch of the stack and jack/wine
> were crashing.
>
> Which version of jack are you using?  Which version of wine?
>
> The last time I tried jack under wine it didn't work but I wasn't able to
> figure out why this time and haven't gone back to it in a few months.
>
> Chris
>
> On Tuesday 21 June 2005 10:02 pm, Alex Villac��s Lasso wrote:
> > Walt Ogburn wrote:
> > > Hi developers,
> > >
> > > I'm trying to understand why the following simple program crashes.  I
> > > think this is the same problem that prevents jack_fst from running with
> > > the current version of wine.  (Jack_fst allows a lot of programs called
> > > VSTs to be run, for audio work).
> > >
> > > I compiled this with:
> > > > winemaker .
> > > > make
> > >
> > > and ran it with:
> > > > wine membug.exe.so
> > >
> > > If BUFSIZE is 1044097, there's an error like this:
> > > err:seh:setup_exception stack overflow 12 bytes in thread 0009 eip
> > > 4057426f esp 40580ff4 stack 0x40580000-0x40680000
> > >
> > > If BUFSIZE is 1044096 or lower, there's no error.  If BUFSIZE is much
> > > higher (like 2000000), there's a segfault.
> > >
> > > All the program does is define a character array of size BUFSIZE and
> > > scribble in it to make sure it really gets allocated.  Maybe this simple
> > > example will show how to make jack_fst work.
> > >
> > > Thanks,
> > > Walter
> > >
> > > ------------------------------------------------------------------------
> > >
> > > #include "windows.h"
> > >
> > > #define BUFSIZE 1044096
> > > /* #define BUFSIZE 2000000 */
> > >
> > > int PASCAL WinMain (HINSTANCE inst, HINSTANCE prev, LPSTR cmdline, int
> > > show) {
> > >   char buf[BUFSIZE];
> > >   int i;
> > >
> > > for (i=0; i<BUFSIZE; i++)
> > >   {
> > >      buf[i] = (char) (i % 256);
> > >   }
> > >
> > > return 0;
> > > }
> >
> > Your sample program is trying to use almost 1 megabyte (!) of stack
> > space. No wonder it crashes. Are you sure that your program jack_fst is
> > trying to allocate such an enormous amount of stack space? Most (sane)
> > programs would just declare a pointer and allocate the required space
> > via malloc() or HeapAlloc().
> >
> > Where can somebody else get a copy of jack_fst? Is there a public
> > download somewhere?
> > Maybe jack_fst corrupts the stack in some strange way that was
> > misdiagnosed as as request for a 1Mb stack?

[Mark Knecht]

Walt,
   (Probably not copied to wine-devel as I'm not subscribed. Pass it
on if you see fit.)

   I haven't played much with this lately, however when I was using it
I saw no real indication that the revision of Jack made any
difference. IIRC I debugged the fix we did down to a certain (huge!)
set of memory management modifications made to Wine all on one day.
Before that set of edits jack_fst worked fine. After that set it never
worked again without your hack to jack_fst. That said I updated Jack
many times after that, probably up through the end of 2004, and never
saw this problem again when using the hack. Since the first of the
year I've hardly touched it and cannot say anything about it at this
time.

   I'm wondering if Chris's note about Jack not working under Wine was
spurred on by my conversations with him a few months ago when I was
getting iTunes working. I tried using Jack with Wine and it didn't
work for me so I contacted Chris but hadn't heard much back since
those conversations.

   Anyway, I think those two (jack_fst and Wine/Jack) are sort of
orthogonal in my mind. I'm not sure the have much to do with each
other but I could be wrong about that. I've tried many times over the
last 2 years to get Wine and Jack to work together and have never been
successful, but jack_fst did work when Jack itself didn't.

   Go figure... ;-)

   If I can be of some help in this let me know.

Cheers,
Mark

On 6/21/05, Walt Ogburn <reuben(a)ugcs.caltech.edu> wrote:
> Thanks Chris.  Mark, do you see the problem go away with a later Jack
> version?  I was just trying to see if I could understand better what was
> going on when we were trying to get this working last year.
>
> - Walter
>
> On Tue, 21 Jun 2005, Chris Morgan wrote:
> > This should have been fixed in jack around 20040325 with a jack to the jack
> > variable, JACK_THREAD_STACK_TOUCH.  The value was changed from 1048576
> > (#define BIG_ENOUGH_STACK in client.c:jack_activate():1033 or so) and moved
> > into configure.in.
> >
> > The change was made in this rev of configure.in:
> > ----------------------------
> > revision 1.253
> > date: 2004/03/26 04:41:57;  author: joq;  state: Exp;  lines: +17 -4
> > [0.96.1] config fixes for shm; stack touch
> > ----------------------------
> >
> > The reason jack does this is to ensure that all pages of the shared memory for
> > all jack clients is mapped so there are no delays during excution.  The value
> > was reduced because wine was taking up a bunch of the stack and jack/wine
> > were crashing.
> >
> > Which version of jack are you using?  Which version of wine?
> >
> > The last time I tried jack under wine it didn't work but I wasn't able to
> > figure out why this time and haven't gone back to it in a few months.
> >
> > Chris
> >
> > On Tuesday 21 June 2005 10:02 pm, Alex Villací­s Lasso wrote:
> > > Walt Ogburn wrote:
> > > > Hi developers,
> > > >
> > > > I'm trying to understand why the following simple program crashes.  I
> > > > think this is the same problem that prevents jack_fst from running with
> > > > the current version of wine.  (Jack_fst allows a lot of programs called
> > > > VSTs to be run, for audio work).
> > > >
> > > > I compiled this with:
> > > > > winemaker .
> > > > > make
> > > >
> > > > and ran it with:
> > > > > wine membug.exe.so
> > > >
> > > > If BUFSIZE is 1044097, there's an error like this:
> > > > err:seh:setup_exception stack overflow 12 bytes in thread 0009 eip
> > > > 4057426f esp 40580ff4 stack 0x40580000-0x40680000
> > > >
> > > > If BUFSIZE is 1044096 or lower, there's no error.  If BUFSIZE is much
> > > > higher (like 2000000), there's a segfault.
> > > >
> > > > All the program does is define a character array of size BUFSIZE and
> > > > scribble in it to make sure it really gets allocated.  Maybe this simple
> > > > example will show how to make jack_fst work.
> > > >
> > > > Thanks,
> > > > Walter
> > > >
> > > > ------------------------------------------------------------------------
> > > >
> > > > #include "windows.h"
> > > >
> > > > #define BUFSIZE 1044096
> > > > /* #define BUFSIZE 2000000 */
> > > >
> > > > int PASCAL WinMain (HINSTANCE inst, HINSTANCE prev, LPSTR cmdline, int
> > > > show) {
> > > >   char buf[BUFSIZE];
> > > >   int i;
> > > >
> > > > for (i=0; i<BUFSIZE; i++)
> > > >   {
> > > >      buf[i] = (char) (i % 256);
> > > >   }
> > > >
> > > > return 0;
> > > > }
> > >
> > > Your sample program is trying to use almost 1 megabyte (!) of stack
> > > space. No wonder it crashes. Are you sure that your program jack_fst is
> > > trying to allocate such an enormous amount of stack space? Most (sane)
> > > programs would just declare a pointer and allocate the required space
> > > via malloc() or HeapAlloc().
> > >
> > > Where can somebody else get a copy of jack_fst? Is there a public
> > > download somewhere?
> > > Maybe jack_fst corrupts the stack in some strange way that was
> > > misdiagnosed as as request for a 1Mb stack?