Re: Is the TSC timecounter safe on SMP system?

John Baldwin wrote:

>On Friday 13 August 2004 01:28 pm, Andrew Gallatin wrote:
>  
>
>>John Baldwin writes:
>> > On Friday 13 August 2004 12:49 pm, Andrew Gallatin wrote:
>> > > I have a system where the TSC timecounter is quite a bit more accurate
>> > > (or perhaps its just much cheaper) than the ACPI timecounter.  This is
>> > > a single CPU, HTT system running an SMP kernel.
>> > >
>> > > A simple program which calls gettimeofday() in a tight loop, looking
>> > > for the microseconds to change sees ~998,000 microsecond updates/sec
>> > > with kern.timecounter.hardware=TSC, and 28,500 updates/sec with
>> > > ACPI-safe.
>> > >
>> > > 1) Is it safe to switch to TSC?
>> > >
>> > > 2) If yes, would it be safe to switch to TSC if this was a real
>> > >    SMP system with multiple physical cpus?
>> >
>> > Probably not.  The problem is that the TSC is not necessarily in sync
>> > between the CPUs so time would "jump around" as you migrated between
>> > CPUs.  If you can get the TSC's synchronized between the CPUs and keep
>> > them that way then you can use the TSC (Linux does this FWIW).
>>
>>But on a single CPU HTT machine, does each HTT core reads the same TSC?
>>    
>>
>
>I think they each have their own TSC.
>
>  
>
Hi,
This is from the Intel system programmer's manual Vol 3
Page 15-56:

There are three ways to count processor clock cycles to monitor 
performance. These are:
• Non-Halted Clockticks – Measures clock cycles in which the specified 
logical processor is
not halted and is not in any power-saving state. When Hyper-Threading 
Technology is
enabled, this these ticks can be measured on a per-logical-processor basis.
• Non-Sleep Clockticks – Measures clock cycles in which the specified 
physical processor is
not in a sleep mode or in a power-saving state. These ticks cannot be 
measured on a
logical-processor basis.
• Time Stamp Counter – Measures clock cycles in which the physical 
processor is not in
deep sleep. These ticks cannot be measured on a logical-processor basis.
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

So I think TSC is not per logical processor but per package.

Thanks
Aniruddha