[ltt-dev] About kernel panic when using Lttng?

[Mathieu Desnoyers]

* jerome zh (jeromezhr at gmail.com) wrote:
> Hi all,
> 
> In normal steps, we have to call the lttctl -D trace1 to stop a trace
> and this makes lttng to write the trace buffer to the trace files.
> So I am wondering how the trace can be stoped if the a kernel panic happens
> and the kernel crashes before calling the lttctl -D trace ?

You'll have to use a tool to extract the trace buffers from the crashed
kernel memory image. Then you can append the last subbuffers at the end
of the tracefiles containing information from the channels already
recorded. Make sure you enable the "LTT_VMCORE" LTTng config option so
you have consistent last subbuffers when the kernel crashes.  You must
use the "commit_seq" counter of the last subbuffers to know how much
contiguous information has been written in the last subbuffer.

I fear that the "crash" tool extentions will not be up to date with the
latest LTTng data structures though.

Thanks,

Mathieu


-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com