Current results: Users of AGP 3.0 (8x) motherboard chipsets often have to address their AGP video cards in PCI mode or force the kernel's GART drivers to try an address their AGP chip in a generic mode. This generic mode may not cleanly work, and can result in system crashes on predictable (and unpredictable?) events. AGP 3.0 support exists in the Linux 2.5 tree. However, it does not exist in the stock 2.4 tree, which is more stable. Desired results: AGP 3.0 chipsets should be supported under at least part of Gentoo's variety of Linux 2.4 kernels. Either a kernel (or several) with the backported AGP 3.0 driver should be made available in ebuild form, or it should be documented that users of AGP 3.0 chipsets need to apply a patch. --- The aforementioned URL leads to a backported patch to the Linux 2.4 AGPGART system to provide AGP 3.0 support in Linux 2.4. While this support exists in the Linux 2.5 tree already (as of 2.5.?), the attached link provides a comfort means for those that really want AGP 3.0 instead of PCI mode but would prefer that the rest of their kernel stay stable. I know from my readings of linux-kernel researching this that Alan Cox is aware of this patch; but I do not know if any of his -ac kernels feature it. Reading his changelog, it seems that he has added a number of GART devices to his kernel; but I personally cannot directly correlate the one Intel device he mentions (the 845G) with the above chip numbers. Someone more C savvy might want to compare his current AGP support with the patch to see if it was properly implemented. This also would provide a low-risk way to close the bug unless someone would like gentoo-sources or similar to contain this support (Note that the original post of this file was corrupt. There is also an accusation of the patch being corrupt again in November 2002, but Alan Cox should be seen countering that it is not. I am personally unaware of any updates to this patch. Someone more kernel savvy and following the lists more should look into this.) Reproducible: Always Steps to Reproduce: This may or may not provide an non-binary-only way to support the NForce(tm)'s chipset with AGP 8x I only know about this issue from work where I have an Intel machine with AGP 3.0 (8X) and this problem.
that would be nice
on the list, but not @ the top. Jay
Dave Jones (2.6 agp maintainer) has said that none of the backports are acceptible to use. In any case 2.6 is nearing the point that it should be useable for most people. Andrew Morton has publicly stated that 2.6.0 will almost certainly be out second half of December.
Yes, but I see no reason we can't backport a 2.5 version if none of the 2.6 "are acceptible for use".
Can I suggest trying out 2.6, now that it is considered the "stable" kernel release?
As an end user, I personally use 2.6 on my one system that needs the AGP 3.0 support since it was released, but this may not be an option in all cases: - Upgrading to Linux 2.6 may break compiling certain programs. For an example, note the various bug reports with various Gnome programs failing to build due to issues with sys-lib/db-1.85 not working properly under Linux 2.6 (Bug #27978 and similar). - The Linux 2.6 Gentoo ebuild (development-sources) still says "PLEASE NOTE THIS IS NOT OFFICIALLY SUPPORTED BY GENTOO" in its pkg_postinst() routine. This suggest (to me, anyway), that the Gentoo team doesn't think 2.6 is ready for primetime. Of course, using backports really is not using "primetime" code either.
I don't see this happening before 2.6 really gets out there and everything else gets fixed up to work well with it. There are currently 2 people on the kernel team that actively work on things, so we can either spend all our time backporting agp from 2.6 or we can do more with what we currently have available. I think we can all agree that it's not a good idea for us to let the current kernels languish for something that is mere moments (in kernel time) away.