On 15 Jan 2003, Michel D�nzer wrote: >> A few people have posted some patches that attempt to make Radeon >> PCI DRI also work on x86, however I don't believe that XFree86 >> sources have applied any of these patches, and I don't believe >> that they've been adequately tested in the wild yet. > >Mike, this is FUD. First of all, PCI GART has been enabled on powerpc as >well for a while, and several people have reported it to work well with >PCI cards on x86.
I enabled Radeon PCI for x86 in Red Hat Linux 7.3. The result was people who did have Radeon PCI filing bug reports, or email/IRC traffic showing that people were experiencing problems. I did not receive any comments from people who did actually get it working on x86, and so I disabled it. Discussions on dri-devel later showed that other people had attempted this as well and it did not work. Some patches came floating around that some people claimed worked. I never tried them, however my recollection was that the patches did not get into CVS and weren't explored much further by anyone. I contacted ATI to get some PCI Radeon hardware, and have planned on experimenting with it. It hasn't been the highest priority to me however, but I did test the Radeon PCI hardware I have out with the 4.2.0 and 4.2.1 releases to no avail. So it most definitely is not FUD. If it worked for as many people as are claiming it does now, then I'd assume it wouldn't be disabled in CVS by default. It'd also likely work on one of the 4 machines I've tested it on - but doesn't. >> Anyone interested in testing out these types of patches need not >> have a Radeon PCI though, you can just disable AGP and an AGP >> card should work as a PCI one for the purposes of testing and >> debugging. > >No, because that seems to be the reason why it's still disabled: it >doesn't work well with AGP cards. Don't ask me why. I can confirm that. I received many bug reports on AGP cards due to this as well, since some users recompile their own kernels, not all users remember or are aware enough to also compile agpgart. That, coupled with a glitch in our installer which caused K5/K6/Cyrix systems to get an i386 kernel with no AGP support, created extra problems with PCIGART enabled. >Anyway, I don't think this is a good reason to keep it disabled >because PCI GART will only be used with Option "ForcePCIMode". >But owners of PCI cards would at least have a chance to try it >without having to modify the source. Instead of giving them the chance... Lets just fix it if we can. ;o) I'd much prefer to fix it. I don't know when I'll have the time to poke at it though, but it is high on my personal todo list, along with other Radeon things. >I've been meaning to rip out the ugly code that disables it for a long >time but sadly haven't gotten around to it yet. I can send you my patch if you like. >> >I've checked the Xfree86.log file and it ends abruptly when the >> >system hangs around the area when the drm extensions(?) are >> >being initialized. If I disable drm in the XF config file the >> >system doesn't hang, but of course the 3D performance sucks, >> >just ask Tuxracer... >> >> Sounds like the problem that occurs on x86 which caused Radeon >> PCI to be disabled by default. > >To me it rather sounds like using an X server where PCI GART is enabled >with a DRM where it isn't. Yeah, could be that I suppose also. Do you have PCI Radeon hardware to fiddle with? -- Mike A. Harris _______________________________________________ XFree86 mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/xfree86
