From Bugzilla Helper: User-Agent: Mozilla/5.0 (compatible; Konqueror/3.2; Linux; en_US, en) (KHTML, like Gecko) Description of problem: In short: the version of voluntary preemption included in the latest testing kernels misses features from the upstream version from Ingo Molnar. This is * voluntary and involuntary preemption can not be enabled at the same time * status of neither can be set in /proc/sys/kernel ++++++++++++++++++++++++++++++++ After reading the article about voluntary preemption at http://kerneltrap.org/node/view/3440 I wanted to try it out. Since Arjan's kernels from http://people.redhat.com/arjanv/2.6/ say that they include voluntary preemption, I downloaded kernel-sourcecode-2.6.7-1.486 and compiled a new kernel from that. (I need my own kernel for read access to NTFS, but that is the only difference to his default kernel) Compiling the kernel showed the first difference to the article: PREEMPT and PREEMPT_VOLUNTARY can not be set at the same time. Afer rebooting I wanted to see what /proc/sys/kernel/voluntary_preemption and /proc/sys/kernel/kernel_preemption say. But neither are available. Looking through patches 1080, 1081, 1082 (from kernel-2.6.7-1.486.src.rpm) it becomes apparent that this version of voluntary preempt is quite a bit different from the one presented at http://people.redhat.com/mingo/voluntary-preempt Is there a reason for that? Version-Release number of selected component (if applicable): kernel-(sourcecode-)2.6.7-1.486 How reproducible: Always Steps to Reproduce: 1.download kernel-2.6.7-1.486.src.rpm 2.rpm -i <file.src.rpm> 3.look at linux-2.6.7-voluntary-preemption.patch, linux-2.6.7-early-schedule.patch, linux-2.6.7-might_sleep.patch in /usr/src/redhat/SOURCES Actual Results: * when recompiling, PREEMPT and PREEMPT_VOLUNTARY can not be set at the same time * knobs in /proc/sys/kernel are missing Expected Results: * when recompiling, PREEMPT and PREEMPT_VOLUNTARY should both be settable * knobs in /proc/sys/kernel are available Additional info: This is not so much a bugreport as a request for comment: why did you cut those features out?
the proc knobs in Ingo's patch were only temporary (as per his announcement); the patch in the kernel rpm is closer to the final form, eg it is always on and far less invasive to the kernel.
Thanks for the information.