To reduce the amount of traces generated by tracers range
and start/stop address filtering has been implemented. This
patch provides example on how to use the various options
currently supported.
Signed-off-by: Mathieu Poirier <mathieu.poirier@xxxxxxxxxx>
Cc: linux-doc@xxxxxxxxxxxxxxx
---
Documentation/trace/coresight.txt | 48 +++++++++++++++++++++++++++++++++++++++
1 file changed, 48 insertions(+)
diff --git a/Documentation/trace/coresight.txt b/Documentation/trace/coresight.txt
index a2e7ccb62c2b..a19d7e60937a 100644
--- a/Documentation/trace/coresight.txt
+++ b/Documentation/trace/coresight.txt
@@ -408,6 +408,54 @@ Instruction 13570831 0x8026B584 E28DD00C false ADD
Instruction 0 0x8026B588 E8BD8000 true LDM sp!,{pc}
Timestamp Timestamp: 17107041535
+Address filtering
+-----------------
+
+CoreSight tracers produce an astonishing amount of traces, more often than not
+much more than trace buffers can handle, and definilty far beyond the
+processing capabilities of humans. To trim down the amount of trace generated
+address filters can be specified from the perf command line.
+
+Currently supported filters include range and start/stop address. When a
+range is specified traces will be collected when the instruction pointer falls
+within the specified addresses. For start and stop filters, traces collection
+start when the instuction pointer matches the start address, and stop when
+the stop address has been reached.
+
+Filters can be specified from the perf command line as follow:
+
+- To start trace collection when the IP equal a specific function in a binary
+ executable or library:
+
+ perf record -e cs_etm// --filter 'start 0x72c@/opt/lib/libcstest.so.1.0' --per-thread ./main
+
+- To stop trace collection when the IP equal a specific function in a binary
+ executable of library:
+
+ perf record -e cs_etm// --filter 'stop 0x790@/opt/lib/libcstest.so.1.0' --per-thread ./main
+
+- To start and stop tracing at various stages of kernel execution:
+
+ perf record -e cs_etm// --filter 'start 0xffffff800856bc50,stop 0xffffff800856bcb0,start 0xffffff8008562d0c,stop 0xffffff8008562d30' --per-thread uname
+
+The only requirement on the number of start/stop filters is that they don't go
+beyond the amount of comparator implemented on the system. Range filters are
+specified using a slightly different syntax:
+
+- To collect traces in a specific range in a binary executable or a library:
+
+ perf record -e cs_etm// --filter 'filter 0x72c/0x40@/opt/lib/libcstest.so.1.0' --per-thread ./main
+
+- To collect traces in a specific range in kernel space:
+
+ perf record -e cs_etm// --filter 'filter 0xffffff8008562d0c/0x48' --per-thread uname
+
+This time the number of comparator _pairs_ implemented in the trace engine is
+the only restriction on the amount of filters that can be specified.
+
+It is possible to mix user and kernel space filters, as long as they are the
+same type. Mixing range and start/stop filters is currently not supported.
+
How to use the STM module
-------------------------
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe linux-doc" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
