Re: [PATCH][3/7] perfctr-2.7.2 for 2.6.6-mm2: x86_64

[John Reiser <jreiser@BitWagon.com>]

>>>+	if( perfctr_cstatus_has_tsc(cstatus) )
>>>+		rdtscl(ctrs->tsc);
>>>+	nrctrs = perfctr_cstatus_nractrs(cstatus);
>>>+	for(i = 0; i < nrctrs; ++i) {
>>>+		unsigned int pmc = state->pmc[i].map;
>>>+		rdpmc_low(pmc, ctrs->pmc[i]);
>>>+	}
>>>
>>>K8 has speculative rdtsc. Most likely you want a sync_core() somewhere
>>>in there.
>>
>>What's the cost for sync_core()? The counts don't have to be
>>perfect.
> 
> 
> It's a CPUID to force a pipeline flush. Let's say 20-30 cycles.

I want the kernel to avoid every delay that can be avoided.  Do not force
a pipeline flush for speculative rdtsc.  Besides those 20-30 cycles
there is register eviction for %eax, %ecx, %edx and save+restore for %ebx
(CPUID scribbles on 4 registers), plus possible branch misprediction if control
is not fall-through sequential.  Also, that kernel code probably is already
several dozen cycles after the most recent user-mode instruction
(the only thing that the user can control), so waiting for quiescent
pipeline is just the kernel lollygagging on itself.  Get to work!

-- 
John Reiser, jreiser@BitWagon.com