On Sat, 2005-01-15 at 11:24 -0500, Michel Dänzer wrote: > On Sat, 2005-01-15 at 12:21 +0100, Jerome Glisse wrote: > > > > I wanted to know if someone with and x86 could test > > if by adding : > > agp_physical=(((get_int(RADEON_MC_AGP_LOCATION))& 0x0ffffU) << 16); > > > > to the end of function void GetMaps(void) all demos > > keep working ? I quite confident that they will :) > > > > The purpose for this is that on PPC (g4 & g5 uninorth & u3) > > agp bridge have some weird bug that make you set agp base > > to 0x0, thus drmAgpBase give you 0x0. Maybe drm lib should be > > fixed and give correct physical agp adress but i do not > > see how to fix this properly in drm (to tell the true haven't > > really looked to fix this in drm :)). > > I think I have too little context here, but FWIW: RADEON_MC_AGP_LOCATION > is the location of the linear AGP aperture as seen by the chip. 0x0 is > probably due to at least UniNorth bridges not providing a physically > linear view of the aperture to the CPU.
No, it is due to another UniNorth bug which force us to put the AGP aperture at address 0 on the bus, though the programs shouldn't care since the radeon it accesses it via it's own mapping at RADEON_MC_AGP_LOCATION (only AGP_BASE_ADDR need to be 0). In addition, as you stated, it's not possible to access the AGP aperture from the CPU (since it overlaps RAM), so a simple ioremap of agp_physical can't work and agpgart mmap won't work neither, though userland apps should have no problem using the DRM to create mappings since it will properly map the RAM pages directly. > > > Now i also get texture looking good with r300_demo, i will > > look at r300_driver see where is the prob. > > Maybe r300_demo doesn't use hostdata blits for uploading the texture > data? > > -- Benjamin Herrenschmidt <[EMAIL PROTECTED]> ------------------------------------------------------- The SF.Net email is sponsored by: Beat the post-holiday blues Get a FREE limited edition SourceForge.net t-shirt from ThinkGeek. It's fun and FREE -- well, almost....http://www.thinkgeek.com/sfshirt -- _______________________________________________ Dri-devel mailing list Dri-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/dri-devel
