Re: Test your agp(4) before blaming on DRM/DRI!

Jung-uk Kim wrote:
> I was trying to make my Intel i815 and VIA K8T800 chipsets working 
> with agp(4) and I found that there were number of issues in the 
> driver.  These issues were mainly from adding PCI IDs blindly without 
> verifying proper documentation.  On top of that, some drivers do not 
> care whether the bridge is supported or not and just attach as 
> 'generic' bridge, e. g., agp_sis.c and agp_via.c.  This gives users 
> false hope and I believe this is wrong.
> 
> I found the following from DRI website:
> 
> http://dri.sourceforge.net/res/testgart.c
> 
> and SuSE Linux had little newer package:
> 
> http://www.suse.de/us/private/products/suse_linux/prof/packages_professional/testgart.html
> 
> I modified SuSE's to work with FreeBSD.  Test your GART before blaming 
> on DRM/DRI!  If aperture size is reported differently from BIOS 
> setting, that is not supported correctly, e. g., i815 - different 
> aperture size encoding and registers.  If writeback test fails, 
> that's more serious problem, e. g., K8T800 - doesn't work at all for 
> me.
> 
> BTW, you have to unload your drm driver (or recompile kernel if you 
> compiled it in) before you test agp.
> 
> I hope this helps many confused users.
> 
> Cheers,
> 
> Jung-uk Kim
> 
> 
> * PS: Is there anybody working on AMD64 on-CPU GART driver like this?
> 
> http://fxr.watson.org/fxr/source/drivers/char/agp/amd64-agp.c?v=linux-2.6.1
> 
> Documentation is here (Section 3.6.11-14):
> 
> http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/26094.PDF

There has been talk about adding support for this into the busdma
framework.  Another nice TODO item if anyone wants it.

Scott