Re: [PATCH 17/17] Add timekeeping documentation

[Zachary Amsden <zamsden@redhat.com>]

On 06/16/2010 10:55 PM, Andi Kleen wrote:
> Zachary Amsden<zamsden@redhat.com>  writes:
>
> I think listing all the obscure bits in the PIT was an attempt to
> weed out the weak and weary readers early, right?
>    

Very perceptive of you ;)

>    
>> +this as well.  Several hardware limitations make the problem worse - if it is
>> +not possible to write the full 32-bits of the TSC, it may be impossible to
>> +match the TSC in newly arriving CPUs to that of the rest of the system,
>> +resulting in unsynchronized TSCs.  This may be done by BIOS or system software,
>> +but in practice, getting a perfectly synchronized TSC will not be possible
>> +unless all values are read from the same clock, which generally only is
>> +possible on single socket systems or those with special hardware
>> +support.
>>      
> That's not true, single crystal for all sockets is very common
> as long as you only have a single motherboard.
>
> Of course there might be other reasons why the TSC is unsynchronized
> (e.g. stop count in C-states), but the single clock is not the problem.
>    

The point is about hotplug CPUs.  Any hotplugged CPU will not have a 
perfectly synchronized TSC, ever, even on a single socket, single 
crystal board.

>    
>> +3.4) TSC and C-states
>> +
>> +C-states, or idling states of the processor, especially C1E and deeper sleep
>> +states may be problematic for TSC as well.  The TSC may stop advancing in such
>> +a state, resulting in a TSC which is behind that of other CPUs when execution
>> +is resumed.  Such CPUs must be detected and flagged by the operating system
>> +based on CPU and chipset identifications.
>> +
>> +The TSC in such a case may be corrected by catching it up to a known external
>> +clocksource.
>>      
> ... This is fixed in recent CPUs ...
>    

And has a CPU flag associated with it (NONSTOP_TSC).  But whether it 
remains fixed across all models and vendors remains to be seen.

>> +
>> +3.5) TSC frequency change / P-states
>> +
>> +To make things slightly more interesting, some CPUs may change requency.  They
>> +may or may not run the TSC at the same rate, and because the frequency change
>> +may be staggered or slewed, at some points in time, the TSC rate may not be
>> +known other than falling within a range of values.  In this case, the TSC will
>> +not be a stable time source, and must be calibrated against a known, stable,
>> +external clock to be a usable source of time.
>> +
>> +Whether the TSC runs at a constant rate or scales with the P-state is model
>> +dependent and must be determined by inspecting CPUID, chipset or various MSR
>> +fields.
>>      
> ... In general newer CPUs should not have problems with this anymore
>    

But that's not the point.  Old CPUs will, and I'm detailing all of the 
existing issues, relevant to new CPUs or not.  A lot of these "old" CPUs 
are still in service and will be for quite some time.

>    
>> +
>> +4) Virtualization Problems
>> +
>> +Timekeeping is especially problematic for virtualization because a number of
>> +challenges arise.  The most obvious problem is that time is now shared between
>> +the host and, potentially, a number of virtual machines.  This happens
>> +naturally on X86 systems when SMM mode is used by the BIOS, but not to such a
>> +degree nor with such frequency.  However, the fact that SMM mode may cause
>>      
> The SMM reference here seems at best odd.
>    

SMIs are notorious for frustrating writers of careful timing loops, and 
several pieces of kernel code take time measurements multiple times to 
rule out outliers from it.

Seems a perfectly reasonable reference to me, perhaps I should explain 
it better.