Skip to content

Commit b72f3fb

Browse files
committed
perf(linux): AM64X: Update PREEMPT_RT perf numbers for 11.1
Update the real-time Linux latency numbers for SDK 11.1 Signed-off-by: Bryan Brattlof <[email protected]>
1 parent 0229a79 commit b72f3fb

File tree

2 files changed

+12
-4
lines changed

2 files changed

+12
-4
lines changed

source/devices/AM64X/linux/RT_Linux_Performance_Guide.rst

Lines changed: 12 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -60,15 +60,23 @@ Test commands used for running stress-ng and cyclictest together
6060
6161
The latencies observed with this DK are summarized below:
6262

63+
.. note::
64+
65+
A known issue in this SDK release is affecting this benchmark.
66+
Applying `this patch`_ on Yocto's meta-ti layer will restore the
67+
context switch latencies to the values you see below.
68+
6369
.. csv-table::
6470
:header: "Latencies","CPU 0","CPU 1"
6571

6672
"Minimum (usec)","6","6"
67-
"Average (usec)","8","8"
68-
"Maximum (usec)","77","105"
73+
"Average (usec)","10","9"
74+
"Maximum (usec)","77","52"
6975

7076
.. image:: img/rt-cpu-method-all-latency-histogram.png
7177

78+
.. _this patch: https://lists.yoctoproject.org/g/meta-ti/message/19089
79+
7280
|
7381
7482
LMBench
@@ -791,8 +799,8 @@ RP Message Inter-Processor Communication (IPC) Latency
791799
RP Message latency Performance
792800
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
793801

794-
RP Message latency is the delay measured from sending a round trip echo message from
795-
a Linux application to a remote processor and back. The following measurements use
802+
RP Message latency is the delay measured from sending a round trip echo message from
803+
a Linux application to a remote processor and back. The following measurements use
796804
a RP message length of 1 byte and 490 bytes for comparison.
797805

798806
The Linux user space application `rpmsg_char_benchmark <https://git.ti.com/cgit/rpmsg/ti-rpmsg-char/>`__ captures these latency values.
355 Bytes
Loading

0 commit comments

Comments
 (0)