SiS drm broken during 2.6.9-rc1-bk1

SiS drm broken during 2.6.9-rc1-bk1

[Dave Airlie]

I've just conducted a binary search to see when the SIS drm stopped
working... and it was 2.6.9-rc1-bk1 from looking at it the new virtual
memory layout is the most likely patch to have broken things... I haven't
confirmed it is this particular patch yet, tomorrow I'll get some time to
do it ..

The SIS is a quite simple drm, so I'm not sure what has been broken, Ingo
can you think of anything that this patch might do to the drm,

the bug is that gears no longer draw anything, but no errors appear
anywhere.. the window is just empty...

Dave.

-- 
David Airlie, Software Engineer
http://www.skynet.ie/~airlied / airlied at skynet.ie
pam_smb / Linux DECstation / Linux VAX / ILUG person

Re: SiS drm broken during 2.6.9-rc1-bk1

[Dave Airlie]

> layout is the most likely patch to have broken things... I haven't
> confirmed it is this particular patch yet, tomorrow I'll get some time to
> do it ..
> 

okay running client applications using 

setarch -L i386 glxgears

makes them work.. I'll start looking for a bug in the the SIS client
side library..

Dave.

Re: SiS drm broken during 2.6.9-rc1-bk1

[Ingo Molnar]

* Dave Airlie <airlied@gmail.com> wrote:

> > layout is the most likely patch to have broken things... I haven't
> > confirmed it is this particular patch yet, tomorrow I'll get some time to
> > do it ..
> > 
> 
> okay running client applications using 
> 
> setarch -L i386 glxgears
> 
> makes them work.. I'll start looking for a bug in the the SIS client
> side library..

yeah. Look for 2GB assumptions - e.g. assumptions that pointers cast to
integer will be positive values, such as:

	int i;

	i = malloc(somesize);
	if (i <= 0)
		handle_alloc_failure();

here with the topdown layout you'd get a malloc 'failure'.

	Ingo

Re: SiS drm broken during 2.6.9-rc1-bk1

[Dave Airlie]

> > > layout is the most likely patch to have broken things... I haven't
> > > confirmed it is this particular patch yet, tomorrow I'll get some time to
> > > do it ..
> > >
> >
> > okay running client applications using
> >
> > setarch -L i386 glxgears
> >
> > makes them work.. I'll start looking for a bug in the the SIS client
> > side library..
> 
> yeah. Look for 2GB assumptions - e.g. assumptions that pointers cast to
> integer will be positive values, such as:
> 
>         int i;
> 
>         i = malloc(somesize);
>         if (i <= 0)
>                 handle_alloc_failure();
> 
> here with the topdown layout you'd get a malloc 'failure'.

Just for completeness, Thomas Winischoffer tracked this down over the
weekend, I had stared at it previously to no great avail,

The issue was with the user space SiS Mesa driver having an
uninitialised structure on the stack for the copy command sent to the
DRM, with the old layout it would end up with zero'ed reserved fields
by luck most of the time, with the new one it went the other way..

the fix is now in Mesa CVS....

Dave.

Re: SiS drm broken during 2.6.9-rc1-bk1

[Ingo Molnar]

* Dave Airlie <airlied@gmail.com> wrote:

> Just for completeness, Thomas Winischoffer tracked this down over the 
> weekend, I had stared at it previously to no great avail,
> 
> The issue was with the user space SiS Mesa driver having an 
> uninitialised structure on the stack for the copy command sent to the 
> DRM, with the old layout it would end up with zero'ed reserved fields 
> by luck most of the time, with the new one it went the other way..

ah, makes sense.

> the fix is now in Mesa CVS....

great!

	Ingo