perf mem: Clarify load-latency in documentation
authorAndi Kleen <ak@linux.intel.com>
Fri, 28 Feb 2014 14:02:14 +0000 (06:02 -0800)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 14 Mar 2014 14:20:44 +0000 (11:20 -0300)
Clarify in the documentation that 'perf mem report' reports use-latency,
not load/store-latency on Intel systems.

This often causes confusion with users.

Signed-off-by: Andi Kleen <ak@linux.intel.com>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/r/1393596135-4227-2-git-send-email-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/Documentation/perf-mem.txt

index 888d51137fbe61e0dc04546b36a46486a9d9a555..1d78a4064da48218b90b15f8495c3626cf614b2c 100644 (file)
@@ -18,6 +18,10 @@ from it, into perf.data. Perf record options are accepted and are passed through
 "perf mem -t <TYPE> report" displays the result. It invokes perf report with the
 right set of options to display a memory access profile.
 
+Note that on Intel systems the memory latency reported is the use-latency,
+not the pure load (or store latency). Use latency includes any pipeline
+queueing delays in addition to the memory subsystem latency.
+
 OPTIONS
 -------
 <command>...::