Description of problem: Talking with tglx on IRC I found that the HR timers is disabled if only a TSC is present. This is because the TSC can not be trusted without having a backup clock to check it against. So if someone configures the kernel correctly, but only enables the TSC, they will be confused to why the HR timers is not enabled. Version-Release number of selected component (if applicable): 2.6.21-rt1 How reproducible: Always Steps to Reproduce: 1. Configure only the TSC clock 2. boot kernel 3. Actual results: No idea why the HR timers is disabled. Expected results: Notify user via bootup message that the HR timers is disabled since only the TSC is enabled. Need a backup clock source to enable the HR timers. Additional info: This should probably be documented somewhere, since I didn't know about this until someone complained, and tglx said that the TSC needs a backup clock to enable HR timers.
Asigning this one to Bhavana, as it likely relates to bug #230638. Minimally this warrants description in our release notes for RHEL-RT.
Summary should probably be "Warn of disabling *HRT*", as other aspects of the -rt patch still should function.
tscsync only enables features in the powernow-k8 driver and does nothing to the Linux timer code. Disable cpufreq and check if using TSC as the time source has the same effect of diabling *HRT*. What you have described might be an artifact of the TSC and not Power Now! + tscsync.
I asked tglx for suggestions on this bug, his reply: 1) we need to add some printk warning, why TSC is not used. We have a mainline patch for this 2) add documentation about it: Requirements for -HRT: - Stable clock source (acpi_pm_timer or hpet) - TSC is not trusted due to: - unsynchronized TSC on SMP - TSC stops in certain power modes (that's probably irrelevant for servers, but we have already seen boxen where the BIOS switches into a deeper power safe idle mode behind the kernels back)
On the above points, I asked tglx for some more specific info. ie, the specific patch and more details for documentation.
Where are we on this bug? Have we actually added printk's to indicate that HRT has been disabled? Clark
As I mentioned in comment #3 a while back this bug is not related to tscsync or Power Now!. Please reassign to a timer expert. Thanks, Bhavana
agreed. reassigning to Thomas
Is this still the case (that HRT is disabled if TSC is only clock source)?