Linux Format - UK (2020-03)

inus Torvalds announced the release

of Linux 5.5-rc7, and implied that it

could be the last before 5.5 (although

he wasn’t sure if he would need an rc8). In any

case, it is certain that Linux 5.5 will be out

imminently. The release of 5.5 will bring the

opening of the “merge window” during which

features intended for 5.6 will be merged.

One of the features we already anticipate

landing in 5.6 will be support for the WireGuard

VPN. WireGuard “aims to be much simpler and

easier to audit than IPsec” and is made

specifically for inclusion into the Linux kernel.

It is only a few thousand lines of code and is

well-supported by distributions. It is also

(apparently) quite easy to use.

Thermal pressure

Modern microprocessors perform (as many

as) billions of calculations every second. The

rate at which these occur depends upon many

factors, but one of them is the frequency of

the processor, or literally how many times

transistors within it switch on and off every

second. Each time a transistor switches, it

consumes a little energy and generates a little

heat. Some additional heat is lost due to static

effects of modern silicon, unrelated to

processor activity.

Processors need to keep their overall heat

(and energy consumption) within certain

envelopes. One of those popularly quoted in

the press (hey that’s us!–Ed) is TDP (thermal

design power), or the amount of heat that the

cooling system can dissipate. In a phone, the

‘cooling system’ is often the case of the phone,

conducting heat away from the chip and into your

pocket. As a result, mobile devices in particular

must operate under some very tight constraints.

The dynamic power consumption of

transistors is a function of voltage squared

times frequency. Consequently, lowering voltage

and frequency has a dramatic effect on power

(and heat). As a result, modern processors use

a feature known as DVFS (dynamic voltage

and frequency scaling) to rapidly make such

adjustments on the fly. Often, this is done to

save energy (and battery life).

When that is the case, the OS scheduler can

usually influence the process, trading power for

performance when applications need instant

responsiveness. Sometimes processors need to

cap performance due to heat. When this happens,

the operating system may have no idea what’s

going on. Enter the Linux Thermal Pressure

patches. Thara Gopinath posted version 7 of a

series intended to teach the scheduler to better

place tasks (applications) in accordance with the

thermal ‘pressure’ seen by a system over time.

Linux will monitor when thermal throttling occurs

and spread out task placement over a range of

processor cores to try to reduce pressure and

regain performance.

On a related note, the ‘IV edition of the Power

Management and Scheduling in the Linux Kernel

(OSPM) summit’ will be in Pisa from 11-13 May.

OPINION

Recently I set up an

OpenStack cluster at

home containing a mix of

x86 and Arm nodes. I used

‘Kolla’ – containerised

OpenStack – and deployed

it using kolla-ansible. This

latter part is supposed to

provide a turnkey solution:

edit a few config files, sit

back and automate the

hundreds of steps involved.

The reality was a little

different. While (eventually)

the scripts more or less

worked, there was a

problem. Guest (‘instance’)

VMs could not reach the

outside world. I traced

various tap, bridge and

other networking interfaces

that connect a guest

through a virtual router and

on to a host network. It led

me to conclude that netfilter

NAT wasn’t working within

the network namespaces

used for the ‘vrouter’ virtual

routers in OpenStack. Many

hours later I realised there

was a bug in the host kernel

on one of my machines.

What I ultimately learned

was that OpenStack is (still)

far too complex for many

‘everyday’ environments.

Kernel debugging is not a

task that average users

should have to be exposed

to. But I was also reminded

how gratifying it is to crawl

up and down the stack for

fun. The pieces are

incredibly complex, but the

source is open, and it is

possible to figure out how it

all goes together.

Jon Masters has been involved

with Linux for more than 22 years.

COMPLEX

OPENSTACK

L

ONGOING DEVELOPMENT

Anup Patel posted v.10 of a patch series

implementing RISC-V support in KVM (the

built-in Linux hypervisor layer). The patches

have been booted successfully under QEMU

and the implementation is pretty complete,

modulo a few pieces of in-kernel emulation

for platform management (SBI v0.2, PLIC,

etc.) that are punted to KVM userspace (in

the host). The patches even try to handle

non-homogeneous RISC-V platforms with

mixed cores, where it might be necessary to

trap and emulate basic features.

Greg Kroah-Hartman published a ‘Linux

Kernel Code of Conduct Committee: August

2019 to November 2019 report’, and then

another for December. He summarised the

number of complaints received under the

Linux Kernel Code of Conduct (www.kernel.

org/code-of-conduct.html). The idea is to

publish anonymised reports to promote civil

discourse upstream. Complaints are handled

through private outreach and coaching.

Linus Torvalds noted that Bruce Evans had

passed away, saying, “Bruce wasn’t ever

really much directly involved in Linux

development – he was active on the BSD

side – but he was the developer behind

Minix/i386, which was what I used for the

original Linux development in the very early

days before Linux became self-hosting”.

Kernel Watch

Jon Masters summarises the latest happenings in

the Linux kernel, so you don’t have to.

10 LXF260 March 2020 http://www.linuxformat.com

NEWSDESK

inusTorvaldsannouncedtherelease

ofLinux5.5-rc7,andimpliedthatit

couldbethelastbefore5.5(although

hewasn’tsureifhewouldneedanrc8).Inany

case,itiscertainthatLinux5.5willbeout

imminently.Thereleaseof5.5willbringthe

openingofthe“mergewindow”duringwhich

featuresintendedfor5.6willbemerged.

Oneofthefeatureswealreadyanticipate

landingin5.6willbesupportfortheWireGuard

VPN.WireGuard“aimstobemuchsimplerand

easiertoauditthanIPsec”andismade

specificallyforinclusionintotheLinuxkernel.

Itisonlyafewthousandlinesofcodeandis

well-supportedbydistributions.Itisalso

(apparently)quiteeasytouse.

Thermalpressure

Modernmicroprocessorsperform(asmany

as)billionsofcalculationseverysecond.The

rateatwhichtheseoccurdependsuponmany

factors,butoneofthemisthefrequencyof

theprocessor,orliterallyhowmanytimes

transistorswithinitswitchonandoffevery

second.Eachtimeatransistorswitches,it

consumesalittleenergyandgeneratesalittle

heat.Someadditionalheatislostduetostatic

effectsofmodernsilicon,unrelatedto

processoractivity.

Processorsneedtokeeptheiroverallheat

(andenergyconsumption)withincertain

envelopes.Oneofthosepopularlyquotedin

thepress(heythat’sus!–Ed)isTDP(thermal

designpower),ortheamountofheatthatthe

coolingsystemcandissipate.Inaphone,the

‘coolingsystem’isoftenthecaseofthephone,

conductingheatawayfromthechipandintoyour

pocket.Asaresult,mobiledevicesinparticular

mustoperateundersomeverytightconstraints.

Thedynamicpowerconsumptionof

transistorsisafunctionofvoltagesquared

timesfrequency.Consequently,loweringvoltage

andfrequencyhasadramaticeffectonpower

(andheat).Asaresult,modernprocessorsuse

afeatureknownasDVFS(dynamicvoltage

andfrequencyscaling)torapidlymakesuch

adjustmentsonthefly.Often,thisisdoneto

saveenergy(andbatterylife).

Whenthatisthecase,theOSschedulercan

usuallyinfluencetheprocess,tradingpowerfor

performancewhenapplicationsneedinstant

responsiveness.Sometimesprocessorsneedto

capperformanceduetoheat.Whenthishappens,

theoperatingsystemmayhavenoideawhat’s

goingon.EntertheLinuxThermalPressure

patches.TharaGopinathpostedversion 7 ofa

seriesintendedtoteachtheschedulertobetter

placetasks(applications)inaccordancewiththe

thermal‘pressure’seenbyasystemovertime.

Linuxwillmonitorwhenthermalthrottlingoccurs

andspreadouttaskplacementoverarangeof

processorcorestotrytoreducepressureand

regainperformance.

Onarelatednote,the‘IVeditionofthePower

ManagementandSchedulingintheLinuxKernel

(OSPM)summit’willbeinPisafrom11-13May.

OPINION

RecentlyIsetupan

OpenStackclusterat

homecontainingamixof

x86andArmnodes.Iused

‘Kolla’–containerised

OpenStack–anddeployed

itusingkolla-ansible.This

latterpartissupposedto

provideaturnkeysolution:

editafewconfigfiles,sit

backandautomatethe

hundredsofstepsinvolved.

Therealitywasalittle

different.While(eventually)

thescriptsmoreorless

worked,therewasa

problem.Guest(‘instance’)

VMscouldnotreachthe

outsideworld.Itraced

varioustap,bridgeand

othernetworkinginterfaces

thatconnectaguest

throughavirtualrouterand

ontoahostnetwork.Itled

metoconcludethatnetfilter

NATwasn’tworkingwithin

thenetworknamespaces

usedforthe‘vrouter’virtual

routersinOpenStack.Many

hourslaterIrealisedthere

wasabuginthehostkernel

ononeofmymachines.

WhatIultimatelylearned

wasthatOpenStackis(still)

fartoocomplexformany

‘everyday’environments.

Kerneldebuggingisnota

taskthataverageusers

shouldhavetobeexposed

to.ButIwasalsoreminded

howgratifyingitistocrawl

upanddownthestackfor

fun.Thepiecesare

incrediblycomplex,butthe

sourceisopen,anditis

possibletofigureouthowit

allgoestogether.

JonMastershasbeeninvolved
withLinuxformorethan 22 years.

COMPLEX

OPENSTACK

L

ONGOINGDEVELOPMENT

AnupPatelpostedv.10ofapatchseries

implementingRISC-VsupportinKVM(the

built-inLinuxhypervisorlayer).Thepatches

havebeenbootedsuccessfullyunderQEMU

andtheimplementationisprettycomplete,

moduloafewpiecesofin-kernelemulation

forplatformmanagement(SBIv0.2,PLIC,

etc.)thatarepuntedtoKVMuserspace(in

thehost).Thepatcheseventrytohandle

non-homogeneousRISC-Vplatformswith

mixedcores,whereitmightbenecessaryto

trapandemulatebasicfeatures.

GregKroah-Hartmanpublisheda‘Linux

KernelCodeofConductCommittee:August

2019 toNovember 2019 report’,andthen

anotherforDecember.Hesummarisedthe

numberofcomplaintsreceivedunderthe

LinuxKernelCodeofConduct(www.kernel.

org/code-of-conduct.html).Theideaisto

publishanonymisedreportstopromotecivil

discourseupstream.Complaintsarehandled

throughprivateoutreachandcoaching.

LinusTorvaldsnotedthatBruceEvanshad

passedaway,saying,“Brucewasn’tever

reallymuchdirectlyinvolvedinLinux

development–hewasactiveontheBSD

side–buthewasthedeveloperbehind

Minix/i386,whichwaswhatIusedforthe

originalLinuxdevelopmentintheveryearly

daysbeforeLinuxbecameself-hosting”.

Kernel Watch

JonMasterssummarisesthelatesthappeningsin

theLinuxkernel,soyoudon’thaveto.

10 LXF260March 2020 http://www.linuxformat.com

NEWSDESK