>>> - In cpuvar.h, I'm not sure if cpu_max_cstates should be added >>> directly to the cpu structure. Is this field going to be used >>> by SPARC? If so, then it's fine. Otherwise, I think it might >>> need to move to the i86pc version of machcpu. >>> >>> >> Yeah, that's a good point and is what I want to ask. >> Will the different idle level be supported by SPARC? >> >> Another issue is, what of C-state need be exported in kstat info? >> Cpu max cstate, latency, power, anything else? >> Any thoughts? >> > > Those are questions better answered by Eric and Bill. But I believe > SPARC just support C0 and C1
SPARC sun4u has a spin idle loop except Olympus/Fujitsu which executes a suspend instruction C1. sun4v calls hv_cpu_yield which I assume enters the hypervisor. Can dtrace probes be placed in the sun4u/sun4v code to track time spent in idle even though idle may not be C1? Bill (Sorry if this is a double post. It looks like my earlier email did not work.)
