Items preceded by [*] means they has to be built into the kernel.
Items preceded by [M] means they has to be built as modules.
Items preceded by [ ] means they should be removed.
- go to the "Instrumentation Support" section
+ go to the "General setup" section
Select the following options :
+ [*] Activate tracepoints
+ [*] Activate markers
+ [*] Activate userspace markers ABI
+ <*> Compile generic tracing probes
+ Linux Trace Toolkit --->
+ [LTTng fine-grained-timestamping]
[*] Linux Trace Toolkit Instrumentation Support
+ <M> or <*> Linux Trace Toolkit Relay+DebugFS Support
+ <M> or <*> Linux Trace Toolkit Serializer
+ <M> or <*> Linux Trace Toolkit Marker Control
<M> or <*> Linux Trace Toolkit Tracer
- <M> or <*> Linux Trace Toolkit Relay+DebugFS Support
- It makes no difference for the rest of the procedure whether the Tracer
- is compiled built-in or as a module.
+ It makes no difference for the rest of the procedure whether the Tracer
+ is compiled built-in or as a module.
activate :
[*] Align Linux Trace Toolkit Traces
- [*] Allow tracing from userspace
<M> Linux Trace Toolkit Netlink Controller
<M> Linux Trace Toolkit State Dump
your choice (see < Help >) :
- [ ] Activate Linux Trace Toolkit Heartbeat Timer
- You may or may not decide to compile probes. Afterward, you will have to
- load the probe modules to enable tracing of their events. The probes
- automatically select the appropriate facilities.
- Static instrumentation is a more invasive type of instrumentation that gives
- the address taking a lock or doing a printk.
+ [ ] Write heartbeat event to shrink traces
+ [ ] Support trace extraction from crash dump
Select <Exit>
Select <Exit>
Select <Yes>
compile statically, you may not need to issue all these commands.
modprobe ltt-control
-modprobe ltt-core
-modprobe ltt-relay
+modprobe ltt-marker-control
modprobe ltt-tracer
-modprobe ltt-probe-mm
-modprobe ltt-probe-kernel
-modprobe ltt-probe-i386 (or x86_64, powerpc, ppc, arm, mips)
-modprobe ltt-probe-net
-modprobe ltt-probe-list
-modprobe ltt-probe-ipc
-modprobe ltt-probe-fs
+modprobe ltt-serialize
+modprobe ltt-relay
+modprobe ipc-trace
+modprobe kernel-trace
+modprobe mm-trace
+modprobe net-trace
+modprobe fs-trace
If you want to have complete information about the kernel state (including all
the process names), you need to load the ltt-statedump module. This is done by
cp /etc/modules /etc/modules.bkp
echo ltt-control >> /etc/modules
-echo ltt-core >> /etc/modules
-echo ltt-relay >> /etc/modules
+echo ltt-marker-control >> /etc/modules
echo ltt-tracer >> /etc/modules
-echo ltt-probe-mm >> /etc/modules
-echo ltt-probe-kernel >> /etc/modules
-echo ltt-probe-i386 >> /etc/modules (or x86_64, powerpc, ppc, arm, mips)
-echo ltt-probe-net >> /etc/modules
-echo ltt-probe-list >> /etc/modules
-echo ltt-probe-ipc >> /etc/modules
-echo ltt-probe-fs >> /etc/modules
-echo ltt-statedump >> /etc/modules
-
-(note : if you want to probe a marker which is within a module, make sure you
-load the probe _after_ the module, otherwise the probe will not be able to
-connect itself to the marker.)
+echo ltt-serialize >> /etc/modules
+echo ltt-relay >> /etc/modules
+echo ipc-trace >> /etc/modules
+echo kernel-trace >> /etc/modules
+echo mm-trace >> /etc/modules
+echo net-trace >> /etc/modules
+echo fs-trace >> /etc/modules
* Getting and installing the ltt-control package (on the traced machine)
make
make install
-* Getting and installing the ltt-usertrace package for user space tracing
-See http://ltt.polymtl.ca/ > USERSPACE TRACING QUICKSTART
+* Getting and installing the markers-userspace package for user space tracing
+See http://ltt.polymtl.ca/packages/markers-userspace-0.5.tar.bz2 or more recent.
* Getting and installing the LTTV package (on the visualisation machine, same or
bottom right label "Current time". Support for this type of filtering should
be added to the filter module soon.
+* Hybrid mode
+
+Starting from LTTng 0.5.105 and ltt-control 0.20, a new mode can be used :
+hybrid. It can be especially useful when studying big workloads on a long period
+of time.
+
+When using this mode, the most important, low rate control information will be
+recorded during all the trace by lttd (i.e. process creation/exit). The high
+rate information (i.e. interrupt/traps/syscall entry/exit) will be kept in a
+flight recorder buffer (now named flight-channelname_X).
+
+The following lttctl commands take an hybrid trace :
+
+Create trace channel, start lttd on normal channels, start tracing:
+lttctl -n tracename -d -l /mnt/relayfs/ltt -t /tmp/trace1 -m hybrid
+
+Stop tracing, start lttd on flight recorder channels, destroy trace channels :
+lttctl -n tracename -f -l /mnt/relayfs/ltt -t /tmp/trace1 -m hybrid
+
+
+We will need to tweak what we consider "important" medium rate events. For
+instance, thread branding events are actually considered a "high rate" event
+when it should be considered "medium rate". The same should apply for the
+state dump process enumeration.
+
+* Flight recorder mode
+
+The flight recorder mode writes data into overwritten buffers for all channels,
+including control channels, except for the facilities tracefiles.
+
+The following lttctl commands take a flight recorder trace :
+
+lttctl -n trace -c -m flight
+lttd -n -d -t /tmp/trace -c /sys/kernel/debug/ltt
+lttctl -n trace -s
+.. do stuff
+lttctl -n trace -q
+lttd -f -d -t /tmp/trace -c /sys/kernel/debug/ltt
+lttctl -m trace -r
+
***********************************************************
** Section 4 * Adding new instrumentations with genevent **
projects / lttv.git / blobdiff