On Mon, 1 Mar 2004, Michel [ISO-8859-1] D�nzer wrote: >> > It looks to me like the error message you're complaining of was not >> > added in 4.3.0. >> >> I don't see anything that is obviously causing this to be triggered. >> However, I am unsure about how this code fits in the wider implementation. > >This is due to the patch 035_tdfx_disable_dri_on_16mb_with_highres.diff, >which contains the comment: > > /* Disable DRI if using a 16Mb card with virtual resolution higher >than > * 1024x768 because DRI does not have enough memory available for >textures > * at higher resolutions, and will not operate correctly. > */ > >See also http://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=38717 . > >Now, this is an odd explanation; 16 MB should be plenty enough for the >DRI at say 1280x1024x16. Mike, do you have a better explanation for this >limit?
It's been quite a long time now since I created that patch. I was receiving bug reports of DRI enabled tdfx configurations crashing randomly when running 3D software, usually video games. The problem didn't usually occur instantly, but would occur over a short period of time, and usually only with rather 3D intensive games like Quake III and the like. When the games were played at a lower resolution, the problems went away. Research into the problem, discovered that it was due to the majority of the video memory being given to the framebuffer, and starving texture memory. I do not remember the gory details, but it should be in dri-devel archives from 1.5-2 years ago. At the time, people proposed that the problem could be solved by modifying the tdfx code to work better in low memory (read as high-resolution) situations. I don't believe anyone ever cared enough about the problem to do that work though, or if it did get done, it certainly has slipped past me unnoticed. ;o) For our (Red Hat) purposes however, I was more concerned with desktop stability than the ability of 3Dfx hardware users to play Quake III at high resolution and crash their machine. My solution was to patch our driver to disable configurations known to cause crashes or other instabilities in certain circumstances, and to remove the patch should the real problem ever be fixed in upstream sources down the line. The patch remains in my current rpm packages for 4.3.0, and I believe many other distributions have borrowed the patch also to prevent the instability problems. Has the tdfx driver been fixed so that memory starvation isn't an issue any more? If so, it's probably a good idea to drop this patch nowadays. I don't have time to do a runtime test personally and bash the driver with some intense 3D. Could someone else do that perhaps? -- Mike A. Harris
