News from the ARM architecture - … from the ARM architecture 1 Historic issues with ARM kernel code 2 Future directions 3 Current workﬂow 4 Ongoing code changes 5 Ideal architecture

c© 2010 IBM Corporation

Linaro ltd.

News from the ARM architecture

Arnd Bergmann


1 Historic issues with ARM kernel code

2 Future directions

3 Current workflow

4 Ongoing code changes

5 Ideal architecture ports

6 Working with the arm-soc process

2 / 47 Arnd Bergmann c© 2010 IBM Corporation


Historic issues withARM kernel code



History of ARM linux code, some time ago

Strong focus on embedded systems

Lots of out of tree codeAging code baseLosing the war on complexityClose to collapse




Strong focus on embedded systemsLots of out of tree code

Aging code baseLosing the war on complexityClose to collapse




Strong focus on embedded systemsLots of out of tree codeAging code base

Losing the war on complexityClose to collapse




Strong focus on embedded systemsLots of out of tree codeAging code baseLosing the war on complexity

Close to collapse




Strong focus on embedded systemsLots of out of tree codeAging code baseLosing the war on complexityClose to collapse



Technical problems in the ARM codebase

Any new hardware mandates code changesNo common platform modelMultiple platforms mutually exclusive at compile timeInfrastructure in hardware specific code



Management problems

Independent vendor specific communitiesLittle incentive for cooperation and peer review

Russell King could no longer keep up⇒ Lack of pushback on crapLinus Torvalds could not keep up either⇒ Lack of merging new code



Management problems

Independent vendor specific communitiesLittle incentive for cooperation and peer reviewRussell King could no longer keep up⇒ Lack of pushback on crap

Linus Torvalds could not keep up either⇒ Lack of merging new code



Management problems

Independent vendor specific communitiesLittle incentive for cooperation and peer reviewRussell King could no longer keep up⇒ Lack of pushback on crapLinus Torvalds could not keep up either⇒ Lack of merging new code



Future directions



Going beyond embedded

DesktopsServers64 bit CPU implementation

One out of the two main linux architectures



Going beyond embedded

DesktopsServers64 bit CPU implementationOne out of the two main linux architectures



Ideal architecture port

Generic user space ABIDevice drivers in subsystems, not subarchitecturesBoot time hardware detectionNo mutually exclusive build time optionsMinimum kernel changes for new hardware



Current workflow



2011 direction changes

Flattened device treeSingle zImage for armv6/v7Duplicate code removalMoving device drivers to subsystemsCo-maintaining arm-soc.git



Changes we did not do

Converge on a single clean platformMove legacy platforms to subdir



2011 achievements

1 new source tree

2 subarch maintainer summits3 merge windows using arm-soc.git4 clean new subarchitectures



2011 achievements

1 new source tree2 subarch maintainer summits

3 merge windows using arm-soc.git4 clean new subarchitectures



2011 achievements

1 new source tree2 subarch maintainer summits3 merge windows using arm-soc.git

4 clean new subarchitectures



2011 achievements

1 new source tree2 subarch maintainer summits3 merge windows using arm-soc.git4 clean new subarchitectures



2012 achievements

Common clock implementationClock DT bindings

Pinctrl subsystemStarted removing board files



2012 achievements

Common clock implementationClock DT bindingsPinctrl subsystem

Started removing board files



2012 achievements

Common clock implementationClock DT bindingsPinctrl subsystemStarted removing board files



Changesets merged

Kernel rmk arm-soc arch/arm total overall2.6.39 680 0 1187 110313.0 265 22 518 98433.1 274 557 980 93803.2 286 701 1355 126953.3 XXX XXX XXX XXX3.4 XXX XXX XXX XXX3.5-rc1 XXX XXX XXX XXX



Top arm-soc contributors by changesets

174 Arnd Bergmann78 Olof Johansson69 Shawn Guo65 Kevin Hilman57 Benoit Cousson53 Kukjin Kim52 Tony Lindgren51 Linus Torvalds50 Jean-Christophe PLAGNIOL-VILLARD46 Fabio Estevam

Total number of contributors: 236



Top contributors by non-merge changesets

69 Shawn Guo65 Kevin Hilman57 Benoit Cousson50 Jean-Christophe PLAGNIOL-VILLARD46 Fabio Estevam43 Stephen Warren43 Mark Brown38 Thomas Abraham34 Paul Walmsley32 Kukjin Kim

Total number of contributors: 235



arm-soc maintainer role

Sponsored by Linaro and GoogleVendor neutralDefine and enforce common rules for everyoneMerging code upstream to LinusCoordination with RussellPushing back on crap



arm-soc workflow

Pulling many per soc topic branchesPushing cross-soc topic branchesOne for-next branch, rebasingEarly staging branchesNot quite clean allowed when doing cleanups

DEMO



arm-soc workflow

Pulling many per soc topic branchesPushing cross-soc topic branchesOne for-next branch, rebasingEarly staging branchesNot quite clean allowed when doing cleanupsDEMO



Ongoing code changes



Multiplatform: Single zImage status

Taking longer than expected

Breakthrough!Progress largely invisibleFocus on ARMv6 and ARMv7Building vexpress+imx+omap2+ux500 nowBooting on vexpress so far




Taking longer than expectedBreakthrough!

Progress largely invisibleFocus on ARMv6 and ARMv7Building vexpress+imx+omap2+ux500 nowBooting on vexpress so far




Taking longer than expectedBreakthrough!Progress largely invisibleFocus on ARMv6 and ARMv7

Building vexpress+imx+omap2+ux500 nowBooting on vexpress so far




Taking longer than expectedBreakthrough!Progress largely invisibleFocus on ARMv6 and ARMv7Building vexpress+imx+omap2+ux500 now

Booting on vexpress so far




Taking longer than expectedBreakthrough!Progress largely invisibleFocus on ARMv6 and ARMv7Building vexpress+imx+omap2+ux500 nowBooting on vexpress so far



Device tree status


Breakthrough!Pinmux bindings in 3.4 (Linus Walleij)Clock bindings in 3.4 (Grant Likely/Mike Turquette)New platforms: prima2, zynq, highbank, picoxcell,spear13xxLargely converted: exynos, i.mx, tegra, at91, vexpress,ux500In progress: omap, shmobile, lpc32xx, msmLots of simple device driver bindings



Device tree status

Taking longer than expectedBreakthrough!

Pinmux bindings in 3.4 (Linus Walleij)Clock bindings in 3.4 (Grant Likely/Mike Turquette)New platforms: prima2, zynq, highbank, picoxcell,spear13xxLargely converted: exynos, i.mx, tegra, at91, vexpress,ux500In progress: omap, shmobile, lpc32xx, msmLots of simple device driver bindings



Device tree status

Taking longer than expectedBreakthrough!Pinmux bindings in 3.4 (Linus Walleij)

Clock bindings in 3.4 (Grant Likely/Mike Turquette)New platforms: prima2, zynq, highbank, picoxcell,spear13xxLargely converted: exynos, i.mx, tegra, at91, vexpress,ux500In progress: omap, shmobile, lpc32xx, msmLots of simple device driver bindings



Device tree status

Taking longer than expectedBreakthrough!Pinmux bindings in 3.4 (Linus Walleij)Clock bindings in 3.4 (Grant Likely/Mike Turquette)

New platforms: prima2, zynq, highbank, picoxcell,spear13xxLargely converted: exynos, i.mx, tegra, at91, vexpress,ux500In progress: omap, shmobile, lpc32xx, msmLots of simple device driver bindings



Device tree status

Taking longer than expectedBreakthrough!Pinmux bindings in 3.4 (Linus Walleij)Clock bindings in 3.4 (Grant Likely/Mike Turquette)New platforms: prima2, zynq, highbank, picoxcell,spear13xxLargely converted: exynos, i.mx, tegra, at91, vexpress,ux500In progress: omap, shmobile, lpc32xx, msm

Lots of simple device driver bindings



Device tree status

Taking longer than expectedBreakthrough!Pinmux bindings in 3.4 (Linus Walleij)Clock bindings in 3.4 (Grant Likely/Mike Turquette)New platforms: prima2, zynq, highbank, picoxcell,spear13xxLargely converted: exynos, i.mx, tegra, at91, vexpress,ux500In progress: omap, shmobile, lpc32xx, msmLots of simple device driver bindings



Randconfig builds


Delayed-by: Arnd Bergmann <[email protected]>

Integration into Linaro CI loopabout 150 patch seriesRegression testingSubmitted by Mathieu Poirier



Randconfig builds

Taking longer than expectedDelayed-by: Arnd Bergmann <[email protected]>

Integration into Linaro CI loop

about 150 patch seriesRegression testingSubmitted by Mathieu Poirier



Randconfig builds

Taking longer than expectedDelayed-by: Arnd Bergmann <[email protected]>

Integration into Linaro CI loopabout 150 patch seriesRegression testingSubmitted by Mathieu Poirier



Memory management changes

Continuous memory allocatordma-buf infrastructuregeneric iommu handling in dma-mapping API



Device driver subsystems to use

MTDDRM (graphics)ASoC (sound)IndustrialIO (ADC, amplifiers, ...)PWMLED



Framework drivers

iommudma-engineregulatorclockgpiolibpinctrl



Platform drivers

irqdomainsparse irqclocksourcedevicetree



Ideal file layout

No board filesDrivers in subsystemsplatform data in include/linux/platform data/register definitions in driversIRQ/GPIO/MMIO/... definitions in DT



Good Examples

HighbankSPEAr13xx



Infrastructure under construction

DMA bindingsSystem controller subsystemdrivers/irqchipearly console abstractionpwm subsystemUSB host controllerSMPPCI I/O space



Cleanup vs. new code development

Reasonable proportion of cleanups

Reducing the amount of crapMore crap getting removed than added

Coding styleNon-portable codeBloatPrivate infrastructureMissing infrastructureExcessive Macro use

Some crap is very well written




Reasonable proportion of cleanupsReducing the amount of crapMore crap getting removed than added

















Platform specific problems: ux500

out of tree codecommon clock




out of tree code

common clock




out of tree codecommon clock



Platform specific problems: omap

hwmodabsolute code size




hwmod

absolute code size




hwmodabsolute code size



Platform specific problems: samsung

fragmentation: exynos, s5p, s3c64xx, s3c24xxpremature DMA DT binding




fragmentation: exynos, s5p, s3c64xx, s3c24xx

premature DMA DT binding




fragmentation: exynos, s5p, s3c64xx, s3c24xxpremature DMA DT binding



Platform specific problems: i.mx

macros in arch/arm/plat-mxc/devices/not much else really




macros in arch/arm/plat-mxc/devices/

not much else really




macros in arch/arm/plat-mxc/devices/not much else really



Platform specific problems: vexpress

clock supportdevice tree support partial




clock support

device tree support partial




clock supportdevice tree support partial



Working with the arm-soc process

Topic branchesSeparate cleanups and bug fixesTiming



Topic branches

Independent developmentBased on a -rc versionCan include dependencies

Too many dependencies:

arm-soc cannot linearize them: Game over

Too many conflicts

cannot resolve: Game over



Topic branches

Independent developmentBased on a -rc versionCan include dependenciesToo many dependencies:

arm-soc cannot linearize them: Game over

Too many conflicts




Topic branches

Independent developmentBased on a -rc versionCan include dependenciesToo many dependencies:

arm-soc cannot linearize them: Game overToo many conflicts




Typical topic branches

fixes (urgent)fixes (non-urgent)cleanupsdtboardpmdrivers



Typical topic branches (continued)

new-socdefconfigmaintainersclockpinctrlnew subsys



Cleanup branches

Usually firstFeatures based on cleanupsLarge but nonintrusive

Your ticket for other changes



Cleanup branches

Usually firstFeatures based on cleanupsLarge but nonintrusiveYour ticket for other changes



Changeset comments

Always too shortExplain why a change is done, not howshortlog style ”ARM: omap: do something”Only bug fixes after -rc1Use ”cc: [email protected]” for backports



Pull requests

Send to [email protected] tagsGet your gpg keys signedMake branches bisectableTag descriptions



Pull timing

As early as possible after -rc1...but no earlier

Staging branchesAs early as possible after -rc1



Summary

Lots of work getting done:device tree, consolidation, cleanupNot running out of work anytime soon:new platforms, ARMv8, multiplatform kernels

Linux on ARM world domination coming,still busy with the details



Summary

Lots of work getting done:device tree, consolidation, cleanupNot running out of work anytime soon:new platforms, ARMv8, multiplatform kernelsLinux on ARM world domination coming,still busy with the details



Legal Statement

This work represents the view of the author and does not necessarilyrepresent the view of IBM.IBM, IBM (logo), e-business (logo), pSeries, e (logo) server, andxSeries are trademarks or registered trademarks of InternationalBusiness Machines Corporation in the United States and/or othercountries.Linux is a registered trademark of Linus Torvalds.Other company, product, and service names may be trademarks orservice marks of others.



Questions?


News from the ARM architecture - … from the ARM architecture 1 Historic issues with ARM kernel code 2 Future directions 3 Current workﬂow 4 Ongoing code changes 5 Ideal architecture

Documents