Skip to content

ELC 2017: Effectively Measure and Reduce Kernel Latencies for Real-time Constraints

License

Notifications You must be signed in to change notification settings

sonicyang/rt-experiments

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ELC2017: Effectively Measure and Reduce Kernel Latencies for Real-time Constraints

This document describe how to install patches we use on linux kernel 4.4 with preempt rt

Grab our patches, configs, etc

Repo: https://github.com/sonicyang/ELC2017 Branch: master

Kernel

Kernel 4.4 patched with PREEMPT_RT

Prepare a source directory of Linux Kernel 4.4 and patch it with PREEMPT_RT

There are kernel patches from mctest, wastedcores, and KML

mctest

Repo: https://github.com/sonicyang/mctest Branch: master In Directory: kernel_patches

WastedCores

Repo: https://github.com/sonicyang/wastedcores Branch: master In Directory: patches

KML

Repo: https://github.com/sonicyang/KML Branch: master In Directory: /

We have updated the KML and Wasted-Cores patches to be compatiable with kernel 4.4. These patches include Kernel-Mode-Linux, vDSO enahncement, the 4 bug-patches, profiler-patch and sanity checker patch. Your can select and patch with the patches your want.

For running mctest, your would have to patch with patches in misc directory.

~/linux-source $ git am ../mctest/kernel_patches/misc/Set-kernel-module-constructor-as-default-enable.patch
~/linux-source $ git am ../mctest/kernel_patches/misc/Floating-point-support-for-kernel-module-constructor.patch

Patching Procedures:

~/linux-source $ git apply <patch file>

Do this for every patches you want to include in your build.

Then, we tar the kernel up for building with buildroot.

~/linux-source $ cd ..
~ $ tar -cvf linux.tar linux-source --exclude='.git'

Buildroot image

Acquire the buildroot repo, we are using version: 2016.11

~ $ git clone git://git.buildroot.net/buildroot
~ $ cd buildroot
~/buildroot $ git checkout 2016.11

Apply Patches to enable building the wastedcore tools, SystemTap, etc.

Procedure:

~/buildroot $ git apply ../ELC2017/buildroot_patches/*.patch

The first patch enable the ability to build systemtap The second enable the ability to build the wastedcore tools, using code in this repo.

I have modified the profiler code as:

  • Enable changing the maxinum recording length and cpu count in buildroot menu.
  • Default:
    • Targeting 4 cores system
    • Shrink the max sched event entry size, total size should be 6 GB
  • Make these configable with CFLAGS
    • -DNUM_CPU
    • -DMAX_SAMPLE_ENTRIES

X86_64

~/buildroot $ cp ../ELC2017/buildroot_configs/x86_64-buildroot-config .config
~/buildroot $ cp ../ELC2017/kernel_configs/x86_64-kernel-config kconfig

ARM Cortex-A9 SOCFPGA

~/buildroot $ cp ../ELC2017/buildroot_configs/ARM-A9-buildroot-config .config
~/buildroot $ cp ../ELC2017/kernel_configs/ARM-A9-kernel-config kconfig
~/buildroot $ cp ../ELC2017/kernel_configs/vmalloc-384M.patch vmalloc-384M.patch

ARM Cortex-A9 SOCFPGA with KML

~/buildroot $ cp ../ELC2017/buildroot_configs/ARM-A9-KML-buildroot-config .config
~/buildroot $ cp ../ELC2017/kernel_configs/ARM-A9-KML-kernel-config kconfig
~/buildroot $ cp ../ELC2017/kernel_configs/vmalloc-384M.patch vmalloc-384M.patch

Configs

Buildroot config

buildroot config is enabled to build image with following packages:

  • rt-tools
  • lmbench
  • systemtap
  • wastecore tools and kernel modules
  • mctest : Coming Soon
~/buildroot $ make menuconfig # Navigate your self to the config entries

Adjust the CPU Number and Max Sample Entries according to your target platform in:

  • Target Packages
    • wastedcores

Kernel config

Kernel config is set to be build with:

  • CONFIG_PREEMPT_RT_FULL

Wastedcore scheduler profiler requires CONFIG_SCHED_DEBUG, and CONFIG_SCHED_INFO

  • CONFIG_SCHED_DEBUG
  • CONFIG_SCHED_INFO

In KML Config, these are enabled:

  • CONFIG_KERNEL_MODE_LINUX=y
  • CONFIG_KML_CHECK_CHROOT=y
  • CONFIG_VDSO=y
  • CONFIG_ARM_LPAE=N

Building Image

~/buildroot $ make

This will start the building process of target image.

mctest might fail to build. It's a bug. Build again without cleaning to over come this

KML user space mctest

# Coming Soon

Install Image

X86_64

Prepare an empty usb disk, the following procedure would erase any data on the drive.

We would dd the output image onto the usb drive, insert your drive and issue:

~/buildroot $ sudo dd if=output/images/disk.img of=/dev/sdx bs=4M
# change sdx according to the driver your using

After a while your driver should be ready for booting

ARM Cortex A9 SOCFPGA

Sign u-boot Image

SOCFPGA requies a signed u-boot. We will get and use mkpinage, made by maximeh, to sign it.

This tool require GO Lang to be presented on your system

~ $ sudo apt-get install golang

Get mkpimage source: https://github.com/maximeh/mkpimage And build it.

~ $ git clone https://github.com/maximeh/mkpimage
~ $ cd mkpimage
~/mkpimage $ export GOPATH=$HOME/go 
~/mkpimage $ make

Sign the u-boot image:

~/buildroot/output/images $ ~/mkpimage/bin/mkpimage u-boot-spl.bin -o u-boot-spl-signed.bin

Prepare SD Card

Create an SD card with following partitioning:

  • Partition 1: 500MB WinFAT(Id: b)
  • Partition 2: 1.5GB Linux(Id:83)
  • Partition 3: 10MB Unknown(Id:a2)

Initialize FAT partition:

# Change sdx according to your drive in system
~/buildroot/output/images $ sudo mkdosfs /dev/sdx1

Install Image to SD Card

Now we can dd the data onto the target SD Card

~/buildroot/output/images $ sudo dd if=u-boot-spl-signed.bin of=/dev/sdx3 bs=64k seek=0
~/buildroot/output/images $ sudo dd if=u-boot.img of=/dev/sdx3 bs=64k seek=4
~/buildroot/output/images $ sudo dd if=rootfs.ext2 of=/dev/sdx2 bs=64k
~/buildroot/output/images $ sudo mkdir /mnt/sdcard
~/buildroot/output/images $ sudo mount /dev/sdx1 /mnt/sdcard
~/buildroot/output/images $ sudo cp socfpga.dtb zImage /mnt/sdcard
~/buildroot/output/images $ sync;sync;sync
# Change sdx according to your drive in system

Your SD card should be ready to boot the Altera SOCFPGA Development Kit!

Boot

Use the newly created driver/SD-Card to boot up the target system.

default user is root, password is empty.

Benchmarking

The testing script we used is located in test_scripts.

The target platform would need to be rebooted every time a test is executed.

Plotting

First copy the result directory, either ouput or output.cyc, as plots/input

~/ELC2017/plots $ sudo cp -r /mnt/<drive>/root/output.cyc input

Heat Map

Then, each test result's heat map can be ploted with:

~/ELC2017/plots $ ~/wastedcores/tools/visualizations_4.4/plots/generate_rows_sched_profiler.sh <BENCH NAME> <Start Second> <Highlighted CTX PID>
# e.g. ~/ELC2017/plots $ ./generate_rows_sched_profiler.sh hackbench.cyc 0 <pid of RT task>

Task running with 's CTX points will be blue, instead of yellow .

Getting 1ms cyclictest's pid

The pid shown in ps command was the main task of cyclictest, not the RT task. To find the RT task, we provided a simple script to find out the pid of tasks running with average of 1ms wakeup time in the profiled data.

First look into input/<BENCH NAME>/<BENCH NAME>.ps, find out the cyclictest main task's PID. We will use it as to filter out other ~1ms tasks.

~/ELC2017/plots $ python get_cyc.py <BENCH NAME> <BASE PID>
# e.g. ~/ELC2017/plots $ python get_cyc.py hackbench.cyc 700

Histogram and other graph for a single PID scheduling result

Requirements:

  • matplotlib

First get the cyclictest RT task PID using above method

Run the python plotters:

~/ELC2017/plots $ python plot_ctx.py <BENCH NAME> <PID>
~/ELC2017/plots $ python plot_cyc.py <BENCH NAME> <PID>
~/ELC2017/plots $ python plot_rq_enter.py <BENCH NAME> <PID>

Plotting result, produced by of any of the tool, would be in output//

Acknowledgements

The patches of Wasted Cores is taken from github repo jirka-h/wastedcores It's ported by user jirka-h from original author, jplozi's repo jplozi/wastedcores, for supporting kernel 4.4.

The KML patches is originated from: http://web.yl.is.s.u-tokyo.ac.jp/~tosh/kml/, developed by: Toshiyuki Maeda Been ported for kernel 4.4 by us.

Additional informations can be find in file, README.md, which is in each patches' directory.

About

ELC 2017: Effectively Measure and Reduce Kernel Latencies for Real-time Constraints

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published