[ros-kernel] Stack Underflow in memory manager (regression)

[Hartmut Birr]

Hi,

> -----Original Message-----
> From: ros-kernel-bounces at reactos.com 
> [mailto:ros-kernel-bounces at reactos.com] On Behalf Of Filip Navara
> Sent: Wednesday, August 25, 2004 10:06 PM
> To: ReactOS Kernel List
> Subject: Re: [ros-kernel] Stack Underflow in memory manager 
> (regression)
> 
> 
> Hi!
> 
> Hartmut Birr wrote:
> 
> >Hi,
> >
> >I've tried to install OpenOffice. If I use the explorer, I 
> must disable line
> >268 in shell32\debughlp.c because the serial line is floated 
> with to many
> >debug message. After starting the setup, the dialog boxes 
> are displayed with
> >a dark blue background and there are printed many debug messages
> >'(objects/color.c:316) NtGdiRealizePalette is unimplemented'.
> >
> That's because DIB pattern brushes are unimplmented. If you change 
> lib/gdi32/misc/stubs.c line 116+ from
>     UNIMPLEMENTED;
>     SetLastError(ERROR_CALL_NOT_IMPLEMENTED);
>     return 0;
> to
>   return NtGdiCreateSolidBrush(0xCCCCCC);
> you'll see some sensible colors.
> 

I've changed back from 16 to 8 bit colors and the colors and some drawing
problems are now ok.  


> >The setup runs
> >up to a message box that oleautobridge.uno.dll is not registerable.
> >
> Hmm, that's after the point of my crash. This error happens because 
> about half of the files isn't extracted for some reason.
> 
> >Mouse
> >moving is very slow. At this point there are open 5000 
> files, there are
> >allocated 4000 cache segments and 40000 rmap entries. Ros 
> does not crash but
> >is not usable. The setup runs on an Athlon XP 1700 with 512MB ram. 

It seems there is an bug in view.c line 284. The trim routine stops if a
page out operation fails. If a page of a segment isn't mapped it does also
fail. The 'break' is wrong on this line. 

> >
> Yes, it *IS* currently  unusable for many reasons and bugs... What 

Unusable means, I can not stop the setup at this point. 

> compilation options have you used? I use DBG=0, 
> KDBG=0,OPTIMIZED=0,ROS_USE_PCH=no,GCC 3.4.1 and latest CVS. 
> This crash 
> wasn't happening few days ago.

I'm using DBG=1, KDBG=0, ROS_USE_PCH=yes (with some modifications for build
and dependency check), GCC 3.4.1 and more optimisation (-O3 for ntoskrnl,
-O6 for some dlls, -mtune=athlon-xp for all). I've also enabled the non
paged pool statistics to get some memory information.

Currently, I try the OpenOffice install again with more debug output. It
hangs at the some point, but it isn't a memory problem because there is
enough memory available:

MC_CACHE 34689, MC_USER 6253, MC_PPOOL 1129, MC_NPPOOL 6667,
MmStats.NrFreePages 81989

- Hartmut