Legacy drivers don't build with recent 2.6.27 kernel. Please apply attached patch, when building on 2.6.27. Reproducible: Always
Created attachment 168310 [details, diff] Compile fix for 2.6.27 kernels
Sorry, attached patch is wrong, because kill_proc_info is not an exported symbol. I'm unable to find an equivalent that works; using kill_pid failes, because the needed find_vpid is a GPL export only which nvidia isn't. I guess we depend on nvidia to be mercyful. :-(
Yup, looks like 2.6.27 requires an updated driver from Nvidia. Keep an eye on http://www.nvnews.net to see when one gets released...
Created attachment 169010 [details, diff] make nvidia-drivers-71.86.06 work with kernel 2.6.27 Following patch in bug #242476 to make nvidia-drivers compile and work. Changes the behaviour by not letting the driver kill procs anymore but nobody knows when the BLOB calls it and how important it is.
Reopen to examine new patch and decide if it's fixworthy.
How well does the driver built with this patch work for you so far? If everything seems normal, it is worth getting the nvidia ebuild maintainer to take a look.
The driver built fine, I'm running 2.6.27 now with accelerated 2D and 3D, I can log out of and can kill the server with ctrl+alt+backspace and everythings seems to work just fine. I don't know when kill_proc is called so I don't know what it would break.
Ok thanks for the patch, let's have the maintainer take a look at it
Thank you, but I have passed maintainership of the nvidia-drivers on to ricmm. I no longer have the appropriate hardware in my laptop.
(In reply to comment #2) > Sorry, attached patch is wrong, because kill_proc_info is not an exported > symbol. I'm unable to find an equivalent that works; using kill_pid failes, > because the needed find_vpid is a GPL export only which nvidia isn't. I guess > we depend on nvidia to be mercyful. :-( > There is an equivalent call on this patch http://bugs.gentoo.org/attachment.cgi?id=170738
w.r.t. the kill_proc modification: I haven't looked at the actual source file, but from the context available it looks like the whole function is being removed, resulting in an undefined value being returned. This would probably cause problems down the line. I would say that "return RM_ERR_OPERATING_SYSTEM;" is probably a suitable replacement (rather than nothing)
Reminder to maintainers: 2.6.27 is planned to go stable on December 15th, it would be great if you could work on this before then
nvidia-drivers-71.86.07 have been added to the tree to support 2.6.27. Please test and report any issues you have with that package.
Marking as fixed since we now have support for 2.6.27.