openbsd's new suspend and resume framework · introduction device conﬁguration activate...

Introduction Device Configuration Activate Functions APM and ACPI Issues Conclusions

OpenBSD’s New Suspend and ResumeFramework

Paul [email protected]

10th European BSD ConferenceOctober 6–9, 2011

Maarssen, The Netherlands


Outline

1 IntroductionHistoryThe Winds of Change

2 Device Configuration3 Activate Functions

ChangesQuiesce

4 APM and ACPIDesignAPMACPI

5 IssuesOverviewQuirks

6 Conclusions


History

Early Days

KISSPower upPower off


History

Time Passes...

Power ManagementComputers start consuming less powerThe system gains some power controlAPM enters the sceneMachines can suspend and resume via APM


The Winds of Change

ACPI

The Machine Gets To Be In ChargeIn theory:

Knob fiddlingBetter controlMore features


The Winds of Change

Implications

Reality CheckExtremely complexSpecifications that nobody respectsEvery vendor has its own quirksThe machine has to do everything


The Winds of Change

Results

ACPINew power management approachAffects device drivers as wellHard to get rightFit into the APM logicLots of system changesNew MI suspend/resume framework


Structure

Kernel Device TrackingTree hierarchyEverything starts at mainbus(4)Device drivers attach to the proper parent device


Example

Dependency viewA memory stick is attached to the system

sd(4)scsibus(4)umass(4)uhub(4)usb(4)ehci(4)pci(4)mainbus(4)

The stick becomes available to the user as sd0.


Configuration

Specific functionalityMatch – proper device driver matchingAttach – attach to a proper place in the device treeActivate – activate the deviceDeactivate – turn off the deviceDetach – remove it from the device tree


Suspend and Resume

Low Power States ImplicationsACPI support required:

New system statesDriver awarenessDevice notification of state changesResult: The need for new activate actions in autoconf(9)


Changes

New Actions

DVACT_QUIESCE

Prepare to suspend (discussed later on).

DVACT_SUSPEND

Set the device drivers in a suspend state.

DVACT_RESUME

Resume the device drivers back to running state.


Changes

Expanding autoconf(9)

Code Changesconfig_suspend()

Similar with attach/detach activate/deactivateSignals the drivers

config_activate_children()

Handle the new casesconfig_suspend() the device’s children


Quiesce

What is Quiesce?

DefinitionThe action of pausing or modifying a given process so that dataconsistency can be achieved.


Quiesce

Why is it important?

Because...Some devices need pre-suspend notifications to:

Finish-up disk I/ODump audio buffersVT switch out of XWait on other actions to finishDo misc. operations requiring a ’normal’ running state


Design

Starting Point

APM MachinesAPM userland daemonUserland notifies the kernelKernel APM MD state machinesLots of MD code, specially for devices


Design

Integrating Other PM Implementations

RulesKeep the same APM mechanism.Mold other implementations into it.Make it opaque to the userland.Let the drivers do the work for them.Implementation specific bits in MDMostly whacky assembly routines


Design

ACPI Implementation

Reiterating...ACPI will be fit in the same modelCreate a fake apm-like kernel ACPI state-machineKeep the same code-paths all the way downNo difference from a userland perspectiveOnly the kernel can tell APM and ACPI apart


APM

Improvements

More MI, Less MDThe BIOS does most of the workRemove MD device related codeLet the device drivers do it in their activate functionsBare MD APM state machine


APM

On Suspend

Code Flowwsdisplay_suspend()

bufq_quiesce()

config_suspend(DVACT_QUIESCE)

splhigh()

disableintr()

config_suspend(DVACT_SUSPEND)

sys_platform->suspend()


APM

On Resume

Code Flowsys_platform->resume()

config_suspend(DVACT_RESUME)

enableintr()

splx()

bufq_restart()

wsdisplay_resume()


ACPI

Implementations

Microsoft WindowsIntel ACPICAOpenBSD


ACPI

How It Works

System PerspectiveACPI is a proxy between the BIOS and the OSAccess AML methods according to the ACPI spec.Lots of spec violationsLots of quirks and workaroundsThe drivers have to handle device state


ACPI

APM-like

FlowThe userland needs no changeacpiioctl() – notfication ioctlSame commands as APMACPI tasks (e.g. acpi_sleep_task())


ACPI

On Suspend

Flowacpi_sleep_task(S3) – checks state changesacpi_sleep_mode(S3) – handles state changesacpi_prepare_sleep_state(S3) – AML nightmareacpi_sleep_machdep(S3) – MD codeacpi_enter_sleep_state(S3) – PM regs fiddling


ACPI

Not APM-Like

AML MethodsTTS – transition to state, before device notificationPTS – prepare to sleep, after device notificationSST – system status indicatorGTS – firmware execution before S3PM – power management registersGTE – wake registers


ACPI

On Resume

Completely different from APMReal-mode: ACPI trampolineReal-mode: Might reenable videoReal-mode: Enable pagingReal-mode: Restore CPU registersJump to where ACPI code stopped during suspendClear PM registersTransition to S0 (more AML methods)Reset the lampEnable runtime GPEsResume the device drivers


Overview

Devices

ProblemsThe order in which we suspend/resume themThe device registersThe memory mapsHow much state do we need to keep?


Overview

ACPI

No Man’s LandThe specifications are just a guide in realityAML is Windows-targetedAML is autogenerated codeMagic methods that poke into CMOS and whatnotThe AML parser is always finding quirks in production code


Quirks

Video

RepostingCan be done by:

Real-mode BIOS callx86emuThe driver itselfNeed for an PCI ID tablenVidia is not supported at allEven then, some cards don’t work


Quirks

USB

ProblemsMost machines have no problems (luck?)Some machines get their usb ports reset on resumeSome don’t get them at allKeep usb state vs whack the whole stack


Quirks

Other

MiscellaneousMount points for usb drives don’t get restoredAudio sometimes gets trashedAucat doesn’t handle suspend/resumeX doesn’t come back on some machinesX gets some image noise, fixed by VT switchingTaking the cpu to 1-cpu is done at the wrong placeSome drivers are not supported yet


Don’t Panic

It Works!Most laptops are supportedMost workstations as wellThe sub-system is stableThe design is goodLoongson is the newest userLots of non suspend/resume bugs in drivers got fixed as aresult


So Long, and Thanks for All the Fish

Questions?

openbsd's new suspend and resume framework · introduction device conﬁguration activate...

Documents