Hi,
I just released LTTng 0.52 for 2.6.27.4. It fixes the stuff I dreamt
about this morning. ;-)
Ingo what do you think about this first point ?
- I now only use 2 bytes to encode the task->state field in the trace.
However it seems that 1 byte would be enough (bitmask of values 0 to
127), except that -1 is used for "unrunnable". But when it's
unrunnable, it should "never" be reach scheduler events, right ?
Changelog :
- Add kernel command line : force_tsc_sync=1 to force tsc
synchronization at boot.
- Supports AMD64 X2 box with unsynchronized TSCs very well. Includes
a warning that tells users to try
force_tsc_sync=1 idle=poll and to disable PowerNow or Speedstep to get
synchronized timestamp counters on x86. This workaround works very
well.
- Default to no LTT alignment for traces if
HAVE_EFFICIENT_UNALIGNED_ACCESS is selected.
- Fixed mixed up irq entry field names in (appeared in 0.51).
Some testing would be required with virtualization to see how it
behaves.
Mathieu
--
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68