Gérer les CPUs sous GNU/Linux

Voir

• Température CPU
• https://www.grid5000.fr/w/CPU_parameters
• https://www.deltasight.fr/processeurs-gestion-energie-frequence/

Outils :

• lscpu
• chcpu
• cpufreq-info
• CoreCtrl
• turbostat
• cpupower
• powertop
• rdmsr / wrmsr (msr-tools)
• kvm-ok / check-bios-nx (cpu-checker)

Notions :

• Systèmes multi-processeurs (NUMA SMP)
• Systèmes multi-cœurs (SMP)
• Systèmes multi-threads (SMT)

Params Kernel :

• nr_cpus

Nombre de threads

grep ^processor /proc/cpuinfo

Nous avons une valeurs par thread. C'est donc la derniere valeur +1 qui donne le nombre de threads (si un seul CPU physique)

Nombre de cores

grep '^cpu cores' /proc/cpuinfo

Toutes les valeurs sont identiques si un seul CPU physique

Voir aussi : intel_p-state

Gérer la fréquence cpu

service cpufrequtils stop

#!/bin/sh -e
#
# rc.local
echo "conservative" >/sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
echo "conservative" >/sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
echo "conservative" >/sys/devices/system/cpu/cpu2/cpufreq/scaling_governor
#echo "conservative" >/sys/devices/system/cpu/cpu3/cpufreq/scaling_governor
# echo "performance" >  /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor
exit 0

Fichier à exécuter en root pour changer le gouverneur. Choix possible selon la configuration du noyau : conservative, ondemand, powersave, performance.

  Avec cpufrequtils : (en root)

cp /usr/share/doc/cpufrequtils/examples/cpufrequtils.sample /etc/default/cpufrequtils
 
cpufreq-set -c 0 -d 2100000 -u 3300000
 
service cpufrequtils reload

1. c : numéro processeur
2. d : freq min
3. u : freq max
4. f : freq exact voulu

Voir :

• https://www.kernel.org/doc/html/v4.14/admin-guide/pm/cpufreq.html
• https://www.kernel.org/doc/Documentation/cpu-freq/user-guide.txt
• https://doc.ubuntu-fr.org/cpu-frequtils
• https://wiki.debian.org/HowTo/CpuFrequencyScaling
• https://wiki.archlinux.org/index.php/CPU_frequency_scaling
• https://www.linuxquestions.org/questions/slackware-14/locking-all-cpu%27s-to-their-maximum-frequency-4175607506/
• https://www.thomas-krenn.com/en/wiki/Disable_CPU_Power_Saving_Management_in_BIOS
• https://bbs.archlinux.org/viewtopic.php?id=163253
• https://doc.ubuntu-fr.org/intel_p-state

grep MHz /proc/cpuinfo

apt-get install linux-cpupower
cpupower

$ cpupower frequency-info -g
analyzing CPU 0:
  available cpufreq governors: Not Available

$ cpupower frequency-info
analyzing CPU 0:
  no or unknown cpufreq driver is active on this CPU
  CPUs which run at the same hardware frequency: Not Available
  CPUs which need to have their frequency coordinated by software: Not Available
  maximum transition latency:  Cannot determine or is not supported.
  hardware limits: Not Available
  available cpufreq governors: Not Available
  Unable to determine current Policy
  current CPU frequency: Unable to call hardware
  current CPU frequency:  Unable to call to kernel
  boost state support:
    Supported: yes
    Active: yes

If you have Turbo Boost enabled in the UEFI/BIOS, you can turn boost on (1) or off (0) by running

echo 1 > /sys/devices/system/cpu/cpufreq/boost

Limite d'utilisation du CPU en %

# cat /sys/devices/system/cpu/intel_pstate/max_perf_pct 
100

cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

Exemble de “gorvernors” possible

• performance
• powersave
• userspace
• ondemand
• conservative

cpufreq-set -r -u 2.24Ghz
#cpufreq-set --cpu 0 --governor conservative
cpufreq-set -r -g performance
 
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_governors 
echo "performance" |tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor 
 
#echo "3900000" > /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq

/etc/default/grub

CMDLINE_LINUX_DEFAULT="quiet splash intel_pstate=disable processor.ignore_ppc=1"

update-grub

Pour remplacer intel_pstate par acpi-cpufreq

cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_driver

La liste des drivers ici :

ls -1 /lib/modules/`uname -r`/kernel/drivers/cpufreq/

Source : https://wiki.archlinux.org/index.php/CPU_frequency_scaling

cat /sys/devices/system/cpu/cpu0/cpufreq/bios_limit

For trying this temporarily change the value in /sys/module/processor/parameters/ignore_ppc from 0 to 1.

echo 1 > /sys/module/processor/parameters/ignore_ppc

For setting it permanent refer to Kernel modules or just read on. Add processor.ignore_ppc=1 to your kernel boot line or create

/etc/modprobe.d/ignore_ppc.conf

# If the frequency of your machine gets wrongly limited by BIOS, this should help
options processor ignore_ppc=1

Avec cpulimit

sudo apt-get install cpulimit

http://www.linuxscrew.com/2012/07/03/limit-cpu-usage-of-linux-process/

perf

apt-get install linux-base linux-tools

https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=4b715d24f4f14731c7b553cbb8604fe865cb8d3c

Voir :

• https://developers.redhat.com/articles/2023/09/29/how-we-solved-hotspot-performance-puzzle

# Disable SMP
echo off > /sys/devices/system/cpu/smt/control
 
# Disable THP
echo "never" > /sys/kernel/mm/transparent_hugepage/enabled
 
# Disable turbo boost
echo 1 > /sys/devices/system/cpu/intel_pstate/no_turbo

Voir :

• https://developers.redhat.com/articles/2023/07/31/frame-pointers-untangling-unwinding?sc_cid=7013a000003Sm66AAC
• https://www.brendangregg.com/perf.html
• https://elinux.org/images/d/d0/Moll--cycle_accurate_profiling_with_perf.pdf

/usr/bin/nproc (coreutils)

getconf _NPROCESSORS_ONLN

service cpufrequtils stop

echo “performance” > /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor

noatime

Swap config

Blacklist module

/proc/interrupts cat /proc/irq/3/smp_affinity

cpupower

http://www.linux-kvm.org/page/CPUHotPlug

• switch to the monitor
• type cpu_set <n+1> online
• A new cpu should appear on /sys/devices/system/cpu
• bring it online by writting 1 to the “online” file inside that directory on guest
• Try adding as many cpus as possible, and make sure we at least do not crash.

pcp http://www.pcp.io/docs/guide.html https://www.youtube.com/watch?v=zrAjevr8_Ds

Real-time kernel ?

http://www.tldp.org/LDP/solrhe/Securing-Optimizing-Linux-RH-Edition-v1.3/chap7sec80.html

https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/6/pdf/Performance_Tuning_Guide/Red_Hat_Enterprise_Linux-6-Performance_Tuning_Guide-en-US.pdf

https://wiki.archlinux.org/index.php/maximizing_performance

https://tweaked.io/guide/kernel/

http://www.redhat.com/promo/summit/2008/downloads/pdf/Wednesday_1015am_John_Shakshober_and_Larry_Woodman_Decoding_the_Code.pdf

http://www.brendangregg.com/blog/2015-03-03/performance-tuning-linux-instances-on-ec2.html

https://oracle-base.com/articles/linux/configuring-huge-pages-for-oracle-on-linux-64

root# dmidecode|grep HTT
HTT (Hyper-Threading Technology)
HTT (Hyper-Threading Technology)

root# egrep 'siblings|cpu cores' /proc/cpuinfo | head -2
siblings : 2
cpu cores : 1

root# grep '^flags\b' /proc/cpuinfo | tail -1
flags : fpu de tsc msr pae mce cx8 apic sep mca cmov pat clflush acpi mmx
fxsr sse sse2 ss ht syscall nx lm constant_tsc rep_good nopl nonstop_tsc pni
pclmulqdq est ssse3 cx16 sse4_1 sse4_2 popcnt aes f16c rdrand hypervisor lahf_lm ida
arat epb pln pts dts fsgsbase erms

root# dmidecode |grep Count # tests that HTT is available *and* enabled
Core Count: 1
Thread Count: 2
Core Count: 1
Thread Count: 2

Source : https://www.oracle.com/docs/tech/ovm-performance.pdf