BIOS Does Not Issue Warm Reset to Work
Around TSC Skew
Issue:
There is a slight chance that the processor Timestamp Counters
(TSC) within each socket will not be synchronized when the system
boots. Newer operating systems that are capable of utilizing the TSC
as a fast high-resolution time source require the TSC to be
synchronized across all sockets to utilize the TSC as a time source.
If the TSCs are not synchronized across sockets, these newer
operating systems will detect this condition and will fall back to a
slower legacy time source. This can result in less than optimal
performance for applications that make heavy use of the
gettimeofday()() system call.
SUSE Linux Enterprise Server (SLES) 11 SP1 is the only supported
operating system affected by this issue.
The error message below reported by SLES 11 SP1 indicates the
operating system has detected that TSCs are not synchronized across
sockets:
[ 5.166336] checking TSC synchronization [CPU#0 ->
CPU#16]:
[ 5.198009] Measured 685 cycles TSC warp between CPUs,
turning off TSC clock.
[ 0.016000] Marking TSC unstable due to
check_tsc_sync_source failed
[ 2.920006] checking TSC synchronization [CPU#0 ->
CPU#17]:
[ 2.924000] Measured 738 cycles TSC warp between CPUs,
turning off TSC clock.
Affected Operating System and
Software:
Workaround:
Reboot the server. The TSCs should be properly synchronized
following the reboot.
|