[ltt-dev] [RFC] LTTng timestamp problems with suspend/resume on x86_64

[Mathieu Desnoyers]

Hi,

I found a problem with LTTng that might be related to the problem
Yannick had when trying to trace a system over suspend and resume on a
x86_64 laptop. It looks like trace data gets recorded while the TSCs are
not in a synchronized state, probably when the system is being brought
back up.

So we probably need to call the in-kernel TSC synchronization code
slightly earlier when the system resumes from sleep. On ARM OMAP3, I use
the external 32k timer which continues running during sleep to make sure
I get a correct timestamp at any point during resume, but on x86, I
wonder if we could use the HPET timer to do the same thing.

Thanks,

Mathieu

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com