On Sat, 05 Jun 2004 14:11:36 +0100, Keith G. Robertson-Turner wrote: > I switched to a stock kernel, in order to reverse the 8K stack patch for > compatibility with nVidia's broken driver. > However, between two different kernels (2.6.5-1.358 and 2.6.6-rc3-bk3) > with nearly identical configs, something has changed to produce a tonne of > garbage being spewed to the console (and /var/log/messages) during bootup. I tried a slightly different approach this time, and it worked! Instead of doing a make rpm on the kernel source, I simply replaced the default i686 config in the SRPM sources with the new one I'd just configured, and did a rpmbuild -ba --target i686 on the original spec file. The resulting kernel still has exactly the same config as before (I ran diffs against a saved copy, /proc/config.gz and /boot/config-`uname -r`), but no verbose audit logging during bootup??? I don't see anything in the original spec file that can account for that - really weird. Also I think I've nailed the specific component responsible, it is the Light-weight Auditing Framework, introduced in 2.6.6, which enables syscall auditing. http://lwn.net/Articles/73623/ The default behaviour is verbose logging, and "turning it off" is a matter of creating a new audit policy. This is all brand new, bleeding-edge stuff, with zero documentation and only very basic userland tools, so people testing 2.6.6 better have plenty of diskspace for their logs, since I can find no way of turning it off (other than disabling syslog). The provided spec applies all the same patches in the rpmbuild, as I do in the manual make rpm build, yet it looks like the result is actually a 2.6.5 kernel rather than 2.6.6-rc3-bk3 - as it it should be if all the patches are applied ??? This is the only explanation I can come up with for why the syscall logging stuff is missing in the spec build, and yet I don't see how it's happening. /me puts on a *very* strong pot of coffee. Anyway, it's ... er ... fixed - I guess. - K.
