Bug 2327

Summary:	AGPGART SIS crashes on sis648
Product:	Drivers	Reporter:	Duarte Santos (dpsantos)
Component:	Video(AGP)	Assignee:	Dave Jones (davej)
Status:	CLOSED PATCH_ALREADY_AVAILABLE
Severity:	blocking	CC:	bmichaelsen
Priority:	P2
Hardware:	i386
OS:	Linux
Kernel Version:	2.6.4 and 2.6.5rc1-mm1	Subsystem:
Regression:	---	Bisected commit-id:
Attachments:	Patch from some guy in kernel mail list A patch against 2.6.5 to let the SiS 746 also get the special treatment

Description Duarte Santos 2004-03-18 08:38:58 UTC

Distribution:GENTOO 2004.0
Hardware Environment: CLEVO LAPTOP P4 2.8 AGP=SIS648 REST=SIS963 DISK=SIS5513
USB=SIS7001 SOUND=SIS7012 GRAPHIC CARD= RADEON M10 9600 
Software Environment: GENTOO - kernel compiled withou acpi and preemp and other
tweaks.
Problem Description: As soon as the agpgart is called it crashes 

Steps to reproduce: Well just compile the kernel as normal with SIS in AGPGART,
i tried as a module too. I also tried severeal ways to boot the pci with bios
and non bios.

Comment 1 Duarte Santos 2004-03-19 10:22:39 UTC

Created attachment 2369 [details]
Patch from some guy in kernel mail list

It's for kernel 2.6.2 .... it works on 2.6.3 too

Comment 2 Bjoern Michaelsen 2004-03-20 10:18:49 UTC

This problem arises also with the SiS 746 agp bridge with AGP 3.0
The system doesnt crash completely though, it can be recovered via SysRq+k or
ssh. A X is running with 99% CPU-time while the machine is unresponsive.
On the second try X starts.
For details see:
http://bugs.gentoo.org/show_bug.cgi?id=40891
This patch:
http://lkml.org/lkml/2004/2/22/102
by Oliver Schoett solved the problem for me, however it seems to be already
included in the patch posted above.
Oliver Schoett on the lkml:
"It appears that the chips are electrically unstable for ca. 5ms after switching
them to AGP 3.0, and the effects vary depending on what your driver does in that
interval."

Comment 3 Duarte Santos 2004-03-25 07:56:22 UTC

Working patch in 2.6.5-rc2-mm3!! Thanks

Comment 4 Bjoern Michaelsen 2004-04-05 19:54:12 UTC

Created attachment 2510 [details]
A patch against 2.6.5 to let the SiS 746 also get the special treatment 

The SiS 746 has the same weird problems during initialisation as the SiS 648.
see my statements from 2004-03-20.
The changes in 2.6.5 make to previous ugly hack superfluous, but apply only to
the SiS 648 and not to the SiS 746.
The new patch against 2.6.5 applies the 5ms wait after initializing AGP to the
SiS 746 too.

Comment 5 Dave Jones 2004-04-06 02:19:17 UTC

thanks, that patch (or something similiar) should already be in the latest -mm
tree. There's quite a few pending agp changesets now, so I'll push those along
when Linus gets back.