Xenomai Kernel Testing

1. Testing script

2. Issues

3. Places to explore

4. Machines

5. Packaging issues

Notes to keep the various machines and problems straight during first tests of the Xenomai kernel packages and the LinuxCNC Xenomai threads port

1. Testing script

To reduce human error while testing, a script to gather system information and run tests is needed.

System information

• dmesg, lspci -vv, /proc/cpuinfo, dmidecode -t 4, lsb_release -a, /proc/devices, lsusb, lsmod, free -m, report-hw, /proc/cmdline
• How to capture HID devices?

Testing

• xeno-regression-test, LinuxCNC latency-test (text version), unit tests

Other tools

• i-pipe tracer http://xenomai.org/index.php/I-pipe:Tracer

2. Issues

CPU hog

• Issue is with idle loop. Keeping CPU at 100% prevents CPU from going into idle loop.
  • Related to latency of transitions of the cpu's idle states
• Reported by skunkworks and Schooner/ArcEye?
• Michael's description on Xeno mailing list http://www.xenomai.org/pipermail/xenomai/2013-February/027675.html
  • Henri Roosen's reply fixes skunkworks: add idle=poll to the kernel command-line
  • Henri also uses nohlt; Jan Kiszka says that will burn power; idle=poll was sufficient for skunkworks

3. Places to explore

• UP kernel packages
• Other deb-like distros: squeeze, lucid, precise
• non-PAE kernels are slow: http://www.xenomai.org/pipermail/xenomai/2013-February/027678.html
• Better control of CPU idle transitions http://biosbits.org/

4. Machines

Schooner's P4 2.4GHz UP

• ich5 chipset
• tested on 3.5.7-xenomai-2.6.2.1
• latency-test
  • RTAI 2.6.24 to 2.6.30, 9k us latency
  • Xenomai 117736 base thread and 87347 servo thread
    • Initial 179000; reset then maxes at above figures
  • particularly sensitive to HID
    • PS/2 mouse and graphics events increased jitter
    • USB mouse did not
• xeno-regr-test: short run; <9us latency

5. Packaging issues

• Michael reports signing key breakage on squeeze