2011-03-17 02:56:28

by Luming Yu

Subject: what is the better method to do tsc synchronized check?

Hi there,

I tested two different methods to do TSC sync check on a Sandy Bridge system.
The first TSC sync check method is from upstream -38-rc that considers
TSC is synced.
The second TSC sync check method is from a 2.6.18 based kernel that
considers TSC
is NOT synced on the Sandy Bridge system. (I modified sync_tsc() to
just test (not modify TSC) on
the cpu being tested.)

[root@intel-sugarbay-DO-01 kernel]# dmesg | grep -i synchroniz
CPU 1: synchronized TSC with CPU 0 (last diff 29 cycles, maxerr 276 cycles)
CPU 2: synchronized TSC with CPU 0 (last diff 32 cycles, maxerr 280 cycles)
CPU 3: synchronized TSC with CPU 0 (last diff 31 cycles, maxerr 284 cycles)

Do you guy think we need to take 2.6.18-based tsc sync testing method
rather than the current one in upstream?

Any inputs are appreciated.


Thanks
Luming

2011-03-17 04:10:07

by Steven Rostedt

Subject: Re: what is the better method to do tsc synchronized check?

[ Added Cc's of people that might help you. ]

-- Steve

On Wed, Mar 16, 2011 at 10:56:25PM -0400, Luming Yu wrote:
> Hi there,
>
> I tested two different methods to do TSC sync check on a Sandy Bridge system.
> The first TSC sync check method is from upstream -38-rc that considers
> TSC is synced.
> The second TSC sync check method is from a 2.6.18 based kernel that
> considers TSC
> is NOT synced on the Sandy Bridge system. (I modified sync_tsc() to
> just test (not modify TSC) on
> the cpu being tested.)
>
> [root@intel-sugarbay-DO-01 kernel]# dmesg | grep -i synchroniz
> CPU 1: synchronized TSC with CPU 0 (last diff 29 cycles, maxerr 276 cycles)
> CPU 2: synchronized TSC with CPU 0 (last diff 32 cycles, maxerr 280 cycles)
> CPU 3: synchronized TSC with CPU 0 (last diff 31 cycles, maxerr 284 cycles)
>
> Do you guy think we need to take 2.6.18-based tsc sync testing method
> rather than the current one in upstream?
>
> Any inputs are appreciated.
>
>
> Thanks
> Luming
> --