Re: [Dri-devel] circular locking in radeon-driver

[Andreas S. <A.S...@gm...>]

1) I modified my local "pre-texmem001merge" dri-sources
to match your patch on current HEAD and it helped. -> OK

2) I started 2 ipers: no real problems (like crashes) here
on my radeon7500.
The problem which remains is, that one process seems to
"get" some of the triangles of the other one.
(This was the issue I was trying to examine and where I stumbled
over the xserver-hang...)
Speculation: Maybe with radeon only some triangles get "stolen"
from one window, but with r200 the chip just crashes?

Dieter: what happens when you run one glxgears and one ipers
at the same time?
what happens with the last pre-texmem-trunk with Michels patch
applied when running 2 ipers?

PS: I am going to try current HEAD as soon as possible.

best regards,
Andreas

Am 2003.05.11 01:21:04 +0200 schrieb(en) Michel D=E4nzer:
> On Sam, 2003-05-10 at 22:21, Dieter N=FCtzel wrote:
> > Am Samstag, 10. Mai 2003 20:44 schrieb Dieter N=FCtzel:
> > > Am Samstag, 10. Mai 2003 12:58 schrieb Michel D=E4nzer:
> > > > On Don, 2003-05-08 at 11:31, Andreas Stenglein wrote:
> >
> > > > > following the ipers_backtrace_3.txt of ipers:
> > > > > #7 radeonGetLock()
> > > > >   gets the lock, then calls
> > > > > #6 radeonUpdatePageFlipping(), wich does a
> > > > >   RADEON_STATECHANGE() in line 77
> > > > >    which does a RADEON_NEWPRIM() in radeon_ioctl.h
> > > > >     which calls rmesa->dma.flush() if it is set.
> > > > >     And it is set: to flush_last_swtcl_prim()....
> > > > >     where we came from, see #10
> > > > >
> > > > > Maybe it would help to copy rmesa->dma.flush to some
> > > > > local functionpointer in RADEON_NEWPRIM(), then
> > > > > set rmesa->dma.flush to 0 and then call the function.
> > > >
> > > > Or simply set rmesa->dma.flush to 0 in flush_last_swtcl_prim()
> before
> > > > calling radeonEmit*() ?
> > >
> > > You are right ;-)
> > >
> > > This one fix it:
>=20
> [...]
>=20
> I just committed something along those lines, thanks for testing.
>=20
>=20
> > But if I start a second (somewhat overlapping) ipers process
> > -> BOOM X server hangs immediately ;-)
> >
> > SYSRQ+"S"/"U" works but NOT "B" (boot).
>=20
> Sounds like another problem. :\
[.....]