Message-ID: <AANLkTi=hoGF7egpyB9sLKJoXv5aiGrxfa9_5iH-96SA=@mail.gmail.com>
Date:	Mon, 2 Aug 2010 21:18:27 -0700
From:	Arve Hjønnevåg <arve@...roid.com>
To:	paulmck@...ux.vnet.ibm.com
Cc:	linux-pm@...ts.linux-foundation.org, linux-kernel@...r.kernel.org,
	mjg59@...f.ucam.org, pavel@....cz, florian@...kler.org,
	rjw@...k.pl, stern@...land.harvard.edu, swetland@...gle.com,
	peterz@...radead.org, tglx@...utronix.de, alan@...rguk.ukuu.org.uk
Subject: Re: Attempted summary of suspend-blockers LKML thread

On Sat, Jul 31, 2010 at 10:58 AM, Paul E. McKenney
<paulmck@...ux.vnet.ibm.com> wrote:
...
> REQUIREMENTS
>
> o       Reduce the system's power consumption in order to (1) extend
>        battery life and (2) preserve state until AC power can be obtained.
>
> o       It is necessary to be able to use power-naive applications.
>        Many of these applications were designed for use in PC platforms
>        where power consumption has historically not been of great
>        concern, due to either (1) the availability of AC power or (2)
>        relatively undemanding laptop battery-lifetime expectations.  The
>        system must be capable of running these power-naive applications
>        without requiring that these applications be modified, and must
>        be capable of reasonable power efficiency even when power-naive
>        applications are available.
>
> o       If the display is powered off, there is no need to run any
>        application whose only effect is to update the display.
>
>        Although one could simply block such an application when it next
>        tries to access the display, it appears that it is highly
>        desirable that the application also be prevented from
>        consuming power computing anything that will not be displayed.
>        Furthermore, whatever mechanism is used must operate on
>        power-naive applications that do not use blocking system calls.
>
> o       In order to avoid overrunning hardware and/or kernel buffers,
>        input events must be delivered to the corresponding application
>        in a timely fashion.  The application might or might not be
>        required to actually process the events in a timely fashion,
>        depending on the specific application.
>
>        In particular, if user input that would prevent the system
>        from entering a low-power state is received while the system is
>        transitioning into a low-power state, the system must transition
>        back out of the low-power state so that it can hand the user
>        input off to the corresponding application.
>
> o       If a power-aware application receives user input, then that
>        application must be given the opportunity to process that
>        input.
>
> o       A power-aware application must be able to efficiently communicate
>        its needs to the system, so that such communication can be
>        performed on hot code paths.  Communication via open() and
>        close() is considered too slow, but communication via ioctl()
>        is acceptable.
>

The problem with using open and close to prevent an allow suspend is
not that it is too slow but that it interferes with collecting stats.
The wakelock code has a sysfs interface that allow you to use a
open/write/close sequence to block or unblock suspend. There is no
limit to the amount of kernel memory that a process can consume with
this interface, so the suspend blocker patchset uses a /dev interface
with ioctls to block or unblock suspend and it destroys the kernel
object when the file descriptor is closed.

> o       Power-naive applications must be prohibited from controlling
>        the system power state.  One acceptable approach is through
>        use of group permissions on a special power-control device.
>
> o       Statistics of the power-control actions taken by power-aware
>        applications must be provided, and must be keyed off of program
>        name.
>

We don't key the stats off the program name, but having useful
statistics is critical too us. The current code in linux-next does not
appear to allow this (I'm referring to pm_stay_awake here, etc not
pm-qos.)

> o       Power-aware applications can make use of power-naive infrastructure.
>        This means that a power-aware application must have some way,
>        whether explicit or implicit, to ensure that any power-naive
>        infrastructure is permitted to run when a power-aware application
>        needs it to run.
>
> o       When a power-aware application is preventing the system from
>        shutting down, and is also waiting on a power-naive application,
>        the power-aware application must set a timeout to handle
>        the possibility that the power-naive application might halt
>        or otherwise fail.  (Such timeouts are also used to limit the
>        number of kernel modifications required.)

wake-lock/suspend-blocker timeouts have nothing to do with the timeout
used by applications when waiting for a response from a less trusted
application.

>
> o       If no power-aware or power-optimized application are indicating
>        a need for the system to remain operating, the system is permitted
>        (even encouraged!) to suspend all execution, even if power-naive
>        applications are runnable.  (This requirement did appear to be
>        somewhat controversial.)

I would say it should suspend even if power aware applications are
runnable. Most applications do not exclusively perform critical work.

>
> o       Transition to low-power state must be efficient.  In particular,
>        methods based on repeated attempts to suspend are considered to
>        be too inefficient to be useful.
>

It must be power-efficient. Repeated attempts to suspend will kill the
idle battery life.

> o       Individual peripherals and CPUs must still use standard
>        power-conservation measures, for example, transitioning CPUs into
>        low-power states on idle and powering down peripheral devices
>        and hardware accelerators that have not been recently used.
>
> o       The API that controls the system power state must be
>        accessible both from Android's Java replacement, from
>        userland C code, and from kernel C code (both process
>        level and irq code, but not NMI handlers).
>
> o       Any initialization of the API that controls the system power
>        state must be unconditional, so as to be free from failure.
>        (I don't currently understand how this relates, probably due to
>        my current insufficient understanding of the proposed patch set.)
>

Unconditional initialization makes it easier to add suspend blockers
to existing kernel code since you don't have to add new failure exit
paths. It is not a strong requirement.

> o       The API that controls the system power state must operate
>        correctly on SMP systems of modest size.  (My guess is that
>        "modest" means up to four CPUs, maybe up to eight CPUs.)
>
> o       Any QoS-based solution must take display and user-input
>        state into account.  In other words, the QoS must be
>        expressed as a function of the display and the user-input
>        states.
>
> o       Transitioning to extremely low power states requires saving
>        and restoring DRAM and/or cache SRAM state, which in itself
>        consumes significant energy.  The power savings must therefore
>        be balanced against the energy consumed in the state
>        transitions.
>
> o       The current Android userspace API must be supported in order
>        to support existing device software.
>
>



-- 
Arve Hjønnevåg
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists